Rework OpenMP thread count limit handling

In order to enable support for future cores which have similar tunings
(in this case I'm doing this for the Arm(R) Neoverse(TM) V2 core), this generically detects SVE support and enables it. This should better manage the size and complexity of dynamic dispatch rather than just copy pasting the same parameters.

To make `ARMV8SVE` more representive of the common 128-bit SVE case,
I've split it and similar parameters from A64FX which has the wider
512-bit SVE.

Propagate SWITCH_RATIO to DYNAMIC_ARCH builds

Previously dynamic builds were either using the default SWITCH_RATIO
or one from the higher level architecture; this patch ensures the
dynamic builds can use this parameter as well.

1. Use UZP instructions but not gather load and scatter store instructions to get lower latency.
    2. Padding k to a power of 4.

Add `OPENBLAS_DEFAULT_NUM_THREADS`

This allows Julia to set a default number of threads (usually `1`) to be
used when no other thread counts are specified [0], to short-circuit the
default OpenBLAS thread initialization routine that spins up a different
number of threads than Julia would otherwise choose.

The reason to add a new environment variable is that we want to be able
to configure OpenBLAS to avoid performing its initial memory
allocation/thread startup, as that can consume significant amounts of
memory, but we still want to be sensitive to legacy codebases that set
things like `OMP_NUM_THREADS` or `GOTOBLAS_NUM_THREADS`.  Creating a new
environment variable that is openblas-specific and is not already
publicly used to control the overall number of threads of programs like
Julia seems to be the best way forward.

[0] https://github.com/JuliaLang/julia/pull/46844

Make it work on general MIPS64 processors

Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com>

It is really dangerous to fallback to Loongson core on other
MIPS64 processors.

Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com>

Some compilers (namely LLVM) are not happy with clobbering
registers in inline assembly.
Use $at as temporary register and explicitly use noat
hint.

Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com>

Add env variable OMP_ADAPTIVE to control OMP threadpool behaviour

It's safer and easier than SIGILL.

Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com>

LoongArch64: Add DYNAMIC_ARCH support