Add dynamic support for Arm(R) Neoverse(TM) V2 processor

[Mousius]

Whilst I figure out how best to map the L2 parameters without duplicating all of ARMV8SVE, lets just map this to NEOVERSEV1.

[martin-frbg]

my thought as well, thanks

[giordano]

$ OPENBLAS_NUM_THREADS=72 OPENBLAS_VERBOSE=2 ./julia -q
Core: armv8sve
julia> peakflops(20_000)
2.0024654326873347e12

julia> using LinearAlgebra

julia> BLAS.lbt_forward("/home/cceamgi/repo/OpenBLAS/libopenblas64_.so"; clear=true)
Core: neoversev1
4860

julia> peakflops(20_000)
2.2361743603730957e12

First peakflops is with OpenBLAS v0.3.26, automatically choosing armv8sve kernels, the second one is after switching to a native build out of this branch (make DYNAMIC_ARCH=1 LIBPREFIX=libopenblas64_ INTERFACE64=1 SYMBOLSUFFIX=64_ -j), which automatically chooses the neoversev1 kernels. The result is consistent with #4440 (comment) where I manually chose the kernels with OPENBLAS_CORETYPE=NEOVERSEV1.