[clang] [llvm] [RFC][AMDGPU] Use `bf16` instead of `i16` for bfloat (PR #80908)

[Shilei Tian via cfe-commits]

shiltian wrote:

The patch is in a good shape now. I have made two other prime patches (#81674 and #81669). I'll rebase this one once they are landed.

This patch only changes one bf16 instruction with the necessary infrastructure for others. I'll update all of them once this patch is landed.

However, I don't think `isInlinableLiteral16` works correctly because the encoding of the floating point inline literals are different for fp16 and bf16, but apparently for now it can only recognize fp16 encoding. This patch at least makes the asm printer work properly. #81345 is trying to fix it correctly, but that is unrelated to this patch.

https://github.com/llvm/llvm-project/pull/80908