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

We were unconditionally linking with libsqlite3 even if OpenCL
was not configured. No other component uses sqlite3 so let's
remove an unneeded dependency when possible.

The development was moved to a fork in the OmpSs-at-FPGA project. For more info visit: https://pm.bsc.es/ompss-at-fpga

This partially reverts commit e5343f02.

The fatal condition is too strict and is affecting several
other scenarios.

The runtime should be able to create as many threads as
available CPUs, or as requested by the user if CPUs are
oversubscribed.

See #1246

Now the execution summary and the nanox-bindings utility print
correctly the bindings mask when one thread is bound to multiple
CPUs.
Also, the string is now more human readable, as binding is grouped
by thread if such is the case.

Fix #1245

Close #1244

Do not fail when the GPU or OpenCL plugin try to reserve an SMP processor but there are not available. Instead, the runtime will show a warning and share the core with another thread

- Capture DLB_Init error code and disable DLB if needed
- Add verbosity to some DLB errors

This option limits the number of threads per core created in SMT machines, while
keeping the same affinity mask by binding to multiple hwthreads inside the core.

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

The main WD should be implicit to honor the implicit parallel
of size 1 that executes the main program.

As a side effect, the main WD in OmpSs can no longer ignore the
barrier construct and must be explicitly prevented.

Recent lock in lendCpu (5cd5ccbe) may cause deadlocks

Dynamic and Guided Worksharings WD submission now only creates
a single WD representative instead of distributing the work.
Idle threads will participate if needed, and will copy the WD
at dequeue time.

Previously, the WD was assuming that the active device in the PE was the first one

Method WD::canRunIn(PE) moved to PE::canRun(WD) because it iterates over all PEdevices, it should be a PE method

Close #1239