The service omp_get_thread_num() now returns -1 if thread is not part of the runtime yet; this query allows to determine either we need to call and admit_current_thread service or not; closes #1247

gcc-8 requires an alias function to have the same exact return type

Old ResourceManager features have been split into different derived
classes in ThreadManager.

The ThreadManager can be specified by a new user option
--thread-manager, which currently accepts 'basic', 'basic-dlb' and
'auto-dlb'. Default value is none

 * Added a cpuProcessMask to keep track of the default process mask
 * Added functions in API, PMInterface and System to access this mask
 * Refactor all the functions to access the cpuActiveMask

By now, this approach is only used by CUDA.

Device specific code moved to plugins, changed the structure
of the System::start to be more generic.
HWLoc, numa info moved to SMPPlugin.
RegionCache: inout copies increase the version on the directory
when doing the copy in (before it was done during the copyDataOut).

Mask services now use a custom type: nanos_cpu_set_t
See #837

This reverts commit 072ac697, reversing
changes made to 9a87a06f.

   * OpenMP apps can now modify their number of threads using
      nanos_omp_*_mask()

It fixes  #801

· There is no threads per PE concept anymore.
· A PE can have 2 threads and another one just one thread.
· SMP threads are binded to their PE's CPU id.
· Added a new parameter, NX_THREADS. OMP_NUM_THREADS is an alias.
· NX_PES only specifies the number of PEs. If you want to run with 4 threads on just 2 PEs, use NX_THREADS=4 and NX_PES=2.
· If NX_THREADS is not specified, it will be set to the number of PEs.
· If NX_PES is not specified, it will use the same number as threads, unless that was also unspecified, in which case the runtime will create as many PEs as the system provides.
· Added System::getMaxThreads() that returns the total number of threads (SMP and GPU). It is used in ayudame and extrae instrumentation.
· System::getCpuId is now const.
· omp_get_num_procs returns the number of available cpus instead of PEs. For instance, if the application is run with NX_PES=2 in a 12-core machine where all of them are available, omp_get_num_procs will return 12. gomp behaves this way.

· There is no threads per PE concept anymore.
· A PE can have 2 threads and another one just one thread.
· SMP threads are binded to their PE's CPU id.
· Added a new parameter, NX_THREADS. OMP_NUM_THREADS is an alias.
· NX_PES only specifies the number of PEs. If you want to run with 4 threads on just 2 PEs, use NX_THREADS=4 and NX_PES=2.
· If NX_THREADS is not specified, it will be set to the number of PEs.
· If NX_PES is not specified, it will use the same number as threads, unless that was also unspecified, in which case the runtime will create as many PEs as the system provides.
· Added System::getMaxThreads() that returns the total number of threads (SMP and GPU). It is used in ayudame and extrae instrumentation.
· System::getCpuId is now const.
· omp_get_num_procs returns the number of available cpus instead of PEs. For instance, if the application is run with NX_PES=2 in a 12-core machine where all of them are available, omp_get_num_procs will return 12. gomp behaves this way.

 extern int  omp_get_level(void);
 extern int  omp_get_ancestor_thread_num(int level);
 extern int  omp_get_team_size(int level);
 extern int  omp_get_active_level(void);
 extern int  omp_in_final(void);

Added simple tests. When parallel is implemented (see #363) these tests will have to be extended.

 * omp_get_num_threads
 * omp_set_num_threads
 * omp_get_thread_num
 * omp_get_max_threads
 * omp_get_nested
 * omp_set_nested
 * omp_get_num_procs
 * omp_in_parallel
 * omp_set_dynamic
 * omp_get_dynamic
 * omp_set_schedule
 * omp_get_schedule
 * omp_get_thread_limit
 * omp_set_max_active_levels
 * omp_get_max_active_levels

They require an OmpData to be associated to WDs which is not yet done.