Fix cooperlake and sapphire rapids march flags on clang

[imciner2]

The march=cooperlake and march=sapphirerapids flags were never getting added when building with Clang targetting those architectures. Instead it was falling back to the skylake AVX512 implementation. This was causing issues when building configurations such as

CC=clang make TARGET=COOPERLAKE BUILD_BFLOAT16=1

because there was either no march on the compiler command, or it was march=skylake-avx512 -mavx512f, neither of which would support the bfloat intrinsics needed, leading to errors like

../kernel/x86_64/stobf16_microk_cooperlake.c:49:20: error: ../kernel/x86_64/sbgemm_tcopy_4_cooperlake.calways_inline function '_mm512_maskz_loadu_ps' requires target feature 'avx512f', but would be inlined into function 'tobf16_accl_kernel' that is compiled without support for 'avx512f':106:21
        __m512 a = _mm512_maskz_loadu_ps(*((__mmask16*) &align_mask16), &in[0]);
                   ^: 

or

In file included from ../kernel/x86_64/tobf16.c:42:
../kernel/x86_64/stobf16_microk_cooperlake.c:46:84: error: always_inline function '_mm512_cvtneps_pbh' requires target feature 'avx512bf16', but would be inlined into function 'tobf16_accl_kernel' that is compiled without support for 'avx512bf16'
        _mm256_mask_storeu_epi16(&out[0], *((__mmask16*) &align_mask16), (__m256i) _mm512_cvtneps_pbh(a));
                                                                                   ^

Clang added support for these two architectures in Clang 9 and Clang 12, so introduce new checks for those versions to enable the appropriate march flag, and fallback to skylake otherwise.

[martin-frbg]

Thanks - seems I completely forgot to push the x86_64 changes when I "fixed" Arm64 compilation with clang recently, and your fix is cleaner anyway.