fix: use x86-64-v3 as target cpu

[usamoi]

Currently releases built by GitHub runners cannot run on machines without AVX-512 instructions. "x86-64-v3" relaxs the restriction for target instruction sets.

[TIAcode]

cargo:warning=./src/f16.c:84:15: error: always_inline function '_mm512_set1_ps' requires target feature 'avx512f', but would be inlined into function 'v_f16_cosine_v4' that is compiled without support for 'avx512f'

The commit that fixed this a year ago isn't included in the sources anymore. Compiling this without AVX-512 runs into a ton of these errors. Would it still be possible to compile without it?

[usamoi]

cargo:warning=./src/f16.c:84:15: error: always_inline function '_mm512_set1_ps' requires target feature 'avx512f', but would be inlined into function 'v_f16_cosine_v4' that is compiled without support for 'avx512f'

The commit that fixed this a year ago isn't included in the sources anymore. Compiling this without AVX-512 runs into a ton of these errors. Would it still be possible to compile without it?

Can you explain what problems do you want to solve? v_f16_cosine_v4 is decorated by __attribute__((target("arch=x86-64-v4,avx512fp16"))) so there should not be this warning.