Add AVX10.2 Intrinsics

[sayantn]

Note: requires rust-lang/rust#153068

• BF16
• Convert
• VNNI
• MinMax
• Saturating Convert
  • Truncating Saturating float -> 32/64 bit integer conversions (cvtts)
  • Saturating float -> 8 bit integer conversions (ipcvt)
• Copy

Cannot link with LLVM (requires rust-lang/rust#140763 or some other patch)

• minmax_pbh
• all of the new BF16 intrinsics
• ipcvt{t}s_bf16_ep{i,u}8

Not Sure: the AVX10.2 document specifies some more instructions, in particular V{U}COMXS{D,S,H}, but I was not able to find any intrinsics for these in GCC or Clang

Note for later: Some intrinsics (specifically VNNI intrinsics and cvtx2ps_ph) don't require the full AVX10.2. They only require the subset enumerated by AVX10_V1_AUX (formerly AVX10_VNNI_INT flag). But neither Clang nor GCC has yet added support for this feature, so change the definitions when they add them.