openblas: bump to 0.3.31

openblas: bump to 0.3.31

Dropped patch: 0001-Make-GEMM3M-parameters-available-on-32bit-X86-GENERI.patch
Part of upstream: https://github.com/OpenMathLib/OpenBLAS/pull/5418

Changelog:
- Revert a 0.3.30 optimization that could cause race conditions
  and invalid results in GEMM
- Fix thread lockup with Python 3.9 and NumPy
- Fix deadlock in multithreaded code after fork()
- Add bfloat16 extensions (BGEMM, BGEMV) and basic FP16 infrastructure
- Add batch GEMM operations with strided variants
- Add multithreaded LAPACK SLAED3/DLAED3 for improved eigensolvers
- Add Apple M4 and Intel Lunar Lake support
- Add initial POWER11 architecture support
- Improve GEMM performance on A64FX and ARM processors

Full release notes:
https://github.com/OpenMathLib/OpenBLAS/releases/tag/v0.3.31

Signed-off-by: Alexandru Ardelean <redacted>