[clang] clang/AMDGPU: Require 16-bit-insts for half typed buffer format builtins (PR #205367)

[Matt Arsenault via cfe-commits]

arsenm wrote:

> We definitely don't need to do the feature check manually.

You do to get the check performed in the correct place in Sema. I would consider relying on the default error in codegen to be a regression and moving the default error a separate problem 

https://github.com/llvm/llvm-project/pull/205367