Improved Arm SVE vectorization backend
Changed the implementation of Arm SVE Simd backend to use the original fixed-size SVE types from SVE ACLE (Chapter 3.7.3.) instead of GCC SIMD types. This has the following benefits:
- work with GNU >= 10, LLVM >= 12 and ARM >= 21.1 compilers
- we can use svbool_t for SimdBool (instead of uint32_t/uint64_t)
- several masking functions and operations are simplified
Edited by Gilles Gouaillardet