Add clang option to avoid running out of registers in AVX512 assembly

* compare denominator against DBL_MIN rather than a somewhat arbitrary small number near it

Fix CMAKE-based cross-compilation to CortexA53 (wrong DGEMM setting)

Observe any stricter (OpenMP) thread count limit imposed by openblas_set_num_threads()

Fix OpenMP thread counting returning places rather than cores

Handle unrecognized ASM compiler (from Arm Compiler 22.1) in CMAKE builds

dot.c early bail fix

Avoid overflow from division in GETF2 potentially causing NaN

Add missing RISC-V architecture in arch.cmake

RISC-V support exists in Makefile.system but is missing in arch.cmake. This patch adds riscv64 platform support to cmake building system just like 039e27545f/Makefile.system (L830-L832) did.

Fix DYNAMIC_ARCH builds that select only a subset of precisions

Fix bug in complex precision tests (c|z)het21 (Reference-LAPACK PR798)

Fix ill-conditioned matrix in LIN testsuite test_rfp (LAPACK PR 796)

Fix double subtraction of N_DEFLATE from istop in ?LAQZ0 (LAPACK 794)

Fix warnings and delete unneeded tests in LAPACKE ?LARFB (LAPACK PR782)

Don't add -tp for the nvc compiler if there is one already in CFLAGS

fix shared and tests prereqs

Fix compiler option setting for AVX512-capable ZEN targets

Add cpuid support for Cortex A715 and X3 by aliasing to A710/X2

Fix USE_PERL option usage

Fix linking to libm  with CMAKE