Add SVE/SVE2 support for ARMv8/ARMv9

[MinetaS]

No functional change

[vondele]

can confirm this compiles cleanly on ARMv9, with correct bench.

I'll be able to look at performance in a couple of days, first quick testing seems neutral.

[MinetaS]

Here is the progress note:

• SVE 2.0 performs worse than SVE 1.0 for unknown reasons
• SVE with 128-bit length registers performs worse than Neon (~1% slowdown, both GCC and Clang)
• Clang is terrible at optimizing SVE code (could have been my mistake, but gcc was ok)

Other remaining changes are trivial and the number below is likely similar to what final version will bring out.

Tested on AWS EC2 c7g.medium (Graviton3, ARM Neoverse V1)

GCC 14.0.1

 speedup        = +0.0333
P(speedup > 0) =  1.0000

Clang 18.1.3

speedup        = -0.0534
P(speedup > 0) =  0.0000

[MinetaS]

Closing, will be introduced later on top of #5543.