[llvm] [mlir] [MLIR][AMDGPU] Adding dynamic size check to avoid subword buffer load (PR #135014)

[Krzysztof Drewniak via llvm-commits]

krzysz00 wrote:

(and also, because these are vector ops, I think negative starting indices might be allowed so that's another scalarization case)

https://github.com/llvm/llvm-project/pull/135014