On OSX 10.12, malloc_default_zone returns a special zone that is not
present in the list of registered zones. That zone uses a "lite zone"
if one is present (apparently enabled when malloc stack logging is
enabled), or the first registered zone otherwise. In practice this
means unless malloc stack logging is enabled, the first registered
zone is the default.

So get the list of zones to get the first one, instead of relying on
malloc_default_zone.

847ff223 added a call to malloc_default_zone() before the main loop in
register_zone, effectively making malloc_default_zone() called twice
without any different outcome expected in the returned result.

It is also called once at the beginning, and a second time at the end
of the loop block.

Instead, call it only once per iteration.

Cray is pretty warning-happy, so disable ones that aren't helpful. Each warning
has a numeric value instead of having named flags to disable specific warnings.
Disable warnings 128 and 1357.

128:  Ignore unreachable code warning. Cray warns about `not_reached()` not
      being reachable in a couple of places because it detects that some loops
      will never terminate.

1357: Ignore warning about redefinition of malloc and friends

With this patch, Cray 8.4.0 and 8.5.1 build cleanly and pass `make check`

Cray uses -herror_on_warning instead of -Werror. Use it everywhere -Werror is
currently used for __attribute__ checks so configure actually detects they're
not supported.

Cray only supports `-M` for generating dependency files. It does not support
`-MM` or `-MT`, so don't try to use them. I just reused the existing mechanism
for turning auto-dependency generation off (`CC_MM=`), but it might be more
principled to add a configure test to check if the compiler supports `-MM` and
`-MT`, instead of manually tracking which compilers don't support those flags.

Get jemalloc building and passing `make check_unit` with cray 8.4. An inlining
bug in 8.4 results in internal errors while trying to build jemalloc. This has
already been reported and fixed for the 8.5 release.

In order to work around the inlining bug, disable gnu compatibility and limit
ipa optimizations.

I copied the msvc compiler check for cray, but note that we perform the test
even if we think we're using gcc because cray pretends to be gcc if `-hgnu`
(which is enabled by default) is used. I couldn't come up with a principled way
to check for the inlining bug, so instead I just checked compiler versions.

The build had lots of warnings I need to address and cray doesn't support -MM
or -MT for dependency tracking, so I had to do `make CC_MM=`.

Add a configure check for __builtin_unreachable instead of basing its
availability on the __GNUC__ version. On OS X using gcc (a real gcc, not the
bundled version that's just a gcc front-end) leads to a linker assertion:

    https://github.com/jemalloc/jemalloc/issues/266

It turns out that this is caused by a gcc bug resulting from the use of
__builtin_unreachable():

    https://gcc.gnu.org/bugzilla/show_bug.cgi?id=57438

To work around this bug, check that __builtin_unreachable() actually works at
configure time, and if it doesn't use abort() instead. The check is based on
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=57438#c21.

With this `make check` passes with a homebrew installed gcc-5 and gcc-6.

The Cray compiler wrappers will often add `-lrt` to the base compiler with
`-static` linking (the default at most sites.) However, `-lrt` isn't
automatically added with `-dynamic`. This means that if jemalloc was built with
`-static`, but then used in a program with `-dynamic` jemalloc won't have
detected that librt is a dependency.

The integration and stress tests use -dynamic, which is causing undefined
references to clock_gettime().

This just adds an extra check for librt (ignoring the autoconf cache) with
`-dynamic` thrown. It also stops filtering librt from the integration tests.

With this `make check` passes for:
 - PrgEnv-gnu
 - PrgEnv-intel
 - PrgEnv-pgi

PrgEnv-cray still needs more work (will be in a separate patch.)

Cray systems come with compiler wrappers to simplify building parallel
applications. CC is the C++ wrapper, and cc is the C wrapper.

The wrappers call the base {Cray, Intel, PGI, or GNU} compiler with vendor
specific flags. The "Programming Environment" (prgenv) that's currently loaded
determines the base compiler. e.g. compiling with gnu looks something like:

    module load PrgEnv-gnu
    cc hello.c

On most systems the wrappers defaults to `-static` mode, which causes them to
only look for static libraries, and not for any dynamic ones (even if the
dynamic version was explicitly listed.)

The integration and stress tests expect to be using the .so, so we have to run
the with -dynamic so that wrapper will find/use the .so.

They are used on all platforms in prng.h.

This builds jemalloc and runs all checks with:
- MSVC 2015 64-bits
- MSVC 2015 32-bits
- MINGW64 (from msys2)
- MINGW32 (from msys2)

Normally, AppVeyor configs are named appveyor.yml, but it is possible to
configure the .yml file name in the AppVeyor project settings such that
the file stays "hidden", like typical travis configs.

Some bug (either in the red-black tree code, or in the pgi compiler) seems to
cause red-black trees to become unbalanced. This issue seems to go away if we
don't use compact red-black trees. Since red-black trees don't seem to be used
much anymore, I opted for what seems to be an easy fix here instead of digging
in and trying to find the root cause of the bug.

Some context in case it's helpful:

I experienced a ton of segfaults while using pgi as Chapel's target compiler
with jemalloc 4.0.4. The little bit of debugging I did pointed me somewhere
deep in red-black tree manipulation, but I didn't get a chance to investigate
further. It looks like 4.2.0 replaced most uses of red-black trees with
pairing-heaps, which seems to avoid whatever bug I was hitting.

However, `make check_unit` was still failing on the rb test, so I figured the
core issue was just being masked. Here's the `make check_unit` failure:

```sh
=== test/unit/rb ===
test_rb_empty: pass
tree_recurse:test/unit/rb.c:90: Failed assertion: (((_Bool) (((uintptr_t) (left_node)->link.rbn_right_red) & ((size_t)1)))) == (false) --> true != false: Node should be black
test_rb_random:test/unit/rb.c:274: Failed assertion: (imbalances) == (0) --> 1 != 0: Tree is unbalanced
tree_recurse:test/unit/rb.c:90: Failed assertion: (((_Bool) (((uintptr_t) (left_node)->link.rbn_right_red) & ((size_t)1)))) == (false) --> true != false: Node should be black
test_rb_random:test/unit/rb.c:274: Failed assertion: (imbalances) == (0) --> 1 != 0: Tree is unbalanced
node_remove:test/unit/rb.c:190: Failed assertion: (imbalances) == (0) --> 2 != 0: Tree is unbalanced
<jemalloc>: test/unit/rb.c:43: Failed assertion: "pathp[-1].cmp < 0"
test/test.sh: line 22: 12926 Aborted
Test harness error
```

While starting to debug I saw the RB_COMPACT option and decided to check if
turning that off resolved the bug. It seems to have fixed it (`make check_unit`
passes and the segfaults under Chapel are gone) so it seems like on okay
work-around. I'd imagine this has performance implications for red-black trees
under pgi, but if they're not going to be used much anymore it's probably not a
big deal.

pgi fails to compile math.c, reporting that `-INFINITY` in `pt_norm_expected[]`
is a "Non-constant" expression. A simplified version of this failure is:

```c
#include <math.h>

static double inf1, inf2 = INFINITY;  // no complaints
static double inf3 = INFINITY;        // suddenly INFINITY is "Non-constant"

int main() { }
```

```sh
PGC-S-0074-Non-constant expression in initializer (t.c: 4)
```

pgi errors on the declaration of inf3, and will compile fine if that line is
removed. I've reported this bug to pgi, but in the meantime I just switched to
using (DBL_MAX + DBL_MAX) to work around this bug.

Revert 245ae603 (Support --with-lg-page
values larger than actual page size.), because it could cause VM map
fragmentation if the kernel grows mmap()ed memory downward.

This resolves #391.

Fix mixed decl in the gettimeofday() branch of nstime_update()

This avoids bootstrapping issues for configurations that require
allocation during tsd initialization.

This resolves #390.

As a side effect this causes the extent's 'committed' flag to be
updated.

Fix a fundamental extent_split_wrapper() bug in an error path.

Fix extent_recycle() to deregister unsplittable extents before leaking
them.

Relax xallocx() test assertions so that unsplittable extents don't cause
test failures.

Deregister extents before deallocation, so that subsequent
reallocation/registration doesn't race with deregistration.

Page-align the gap, if any, and add/use extent_dalloc_gap(), which
registers the gap extent before deallocation.

Now that extents are not multiples of chunksize, it's necessary to track
pages rather than chunks.

With the removal of subchunk size class infrastructure, there are no
large size classes that are guaranteed to be re-expandable in place
unless munmap() is disabled.  Work around these legitimate failures with
rallocx() fallback calls.  If there were no test configuration for which
the xallocx() calls succeeded, it would be important to override the
extent hooks for testing purposes, but by default these tests don't use
the rallocx() fallbacks on Linux, so test coverage is still sufficient.

This facilitates the application accessing its own extent allocator
metadata during hook invocations.

This resolves #259.

rtree-based extent lookups remain more expensive than chunk-based run
lookups, but with this optimization the fast path slowdown is ~3 CPU
cycles per metadata lookup (on Intel Core i7-4980HQ), versus ~11 cycles
prior.  The path caching speedup tends to degrade gracefully unless
allocated memory is spread far apart (as is the case when using a
mixture of sbrk() and mmap()).

rallocx() for an alignment-constrained request may end up with a
smaller-than-worst-case size if in-place reallocation succeeds due to
serendipitous alignment.  In such cases, sampling may not happen.

In the case where prof_alloc_prep() is called with an over-estimate of
allocation size, and sampling doesn't end up being triggered, the tctx
must be discarded.