Allow building on OpenBSD

With this change, OpenBLAS builds and all tests pass on OpenBSD 6.2
using Clang. Tested on x86-64 only, with and without DYNAMIC_ARCH=1.

Old gcc and clang do not handle constructor arguments, finally fix #875 as discussed there, using the fedora patch

to migitate performance problems caused by #1052 and #1299 as seen in #1461

Even if we're directly using the win32 threading driver and not pthreads,
pthread_atfork still works fine to register a pre-fork handler, and is
necessary to restore the threading server to a pre-initialized state.

Initialize unitialized values for repeated calls

Address warnings #1357

Address few more warnings

* Calculation of chunk range limits was ignoring num_cpu

bug introduced by me in #1262 - should fix #1388

* Calculation of range limits was ignoring num_cpu

bug introduced by me in #1262

* Calculation of chunk range limits was ignoring num_cpu

bug introduced by me in #1262

* Calculation of chunk range limits was ignoring num_cpu

bug introduced by me in #1262

* Calculation of chunk range limits was ignoring num_cpu

bug introduced by me in #1262

* Calculation of chunk range limits was ignoring num_cpu

bug introduced by me in #1262

* dead increments in driver/level2

* dead increments in kernel/generic

* part dead increments in kernel/x86_64

Work around errors in multithreaded dtrmv

Some Linux distributions (notably SuSE) have raised CPU_SETSIZE to 4096, apparently disregarding API limitations.
From #1348, the highest value to survive array initialization (on a desktop system) is 3232, and 1024 - which is the 
more usual CPU_SETSIZE limit, was demonstrated to work fine on an actual bignuma system.

as a (hopefully temporary) workaround for #1332

Failures to obtain or attach shared memory segments would lead to an exit without explanation of the exact cause.
This change introduces a more verbose error message and tries to make the code continue without setting cpu affinity.
Fixes #1351

also replaces one verbatim sched_yield with the YIELDING macro for consistency as suggested in #1351

2D thread distribution for multi-threaded GEMMs

CMake improvements

Allows maximum use of available cores if one of M and N is small and the other is large.

... as the "allocation" code jumped to now does UNLOCK_COMMAND instead of blas_unlock

Make sure that range limit of last thread never exceeds data size

CMake Improvements

Rework __GLIBC_PREREQ checks to avoid breaking non-glibc builds