Previously this was left without `SWITCH_RATIO` or `GEMM_PREFERED_SIZE` and older default values, but it can be seen across other cores that these values seem to work for many devices.

This allows us to use the `ARMV8` and `ARMV8SVE` targets in a minimal build without losing too much performance.

Forward GEMM to GEMV when one argument is actually a vector

* handle INF and NAN in input (for S/D only if DUMMY2 argument is set)

This helps to reduce values going missing as we accumulate.

Continue requesting ubuntu18 instead of latest on OSUOSL powerCI

Fix GCC11 check for A64FX target

Add fallback compile options for A64FX target

Fix c_check to tolerate a dashed suffix to the gcc version number

Re-enable the SGESDD benchmark after the SCAL fixes

[Docs] replace "Preview" in the MSVC vcvarsall path example

Fix SCAL on x86 and RISCV_GENERIC

[WIP]Make NAN handling in the SCAL kernels depend on the dummy2 parameter

Try to fixed numpy ci test failures

Small GEMM for AArch64 with SVE

Add A64FX to the list of CPUs supported by DYNAMIC_ARCH

Fix incompatible definitions of MAXLOC in f2c-converted LAPACK sources

Work around a gcc14.1 bug that breaks utest on Loongarch

Add autodetection for riscv64

Change computational order in GEMV and GEMM Power6 kernel