[MLIR][LLVMIR] Add 'align' to loop-annotations (#234263)
This is necessary for the llvm.loop.align flag, which clang needs to
generate. Its a touch weird in that LLVM's verifier (not the MLIR
dialect!) enforces this as a 32 bit value, so we did have to do a
little work to make sure we enforce that too/sooner in the dialect.
