Added AVX512BW support

[Marabii]

Hello maintainers,

I added AVX512BW support, a new implementation next to the others (avx2, ...)
It is automatically selected in runtime as the fastest solution when available just like before.

flatten_bits uses AVX-512VBMI2's vpcompressb when available, with a
scalar fallback when it isn't.

I had to bump up SIMDJSON_PADDING from 32 to 64 to match the __m512i width.

MSRV bumped to 1.89 (required by the AVX-512 intrinsics), CI toolchains and README updated accordingly.

Benchmarked against AVX2 on a 550MB generated JSON file: ~5% faster on to_tape and roughly on par for full to_value, since that path is dominated by hashmap/allocation costs rather than stage-1 parsing

Full test suite ran perfectly.

Please let me know if there are any modifications needed before it can be merged.

Thanks!