[clang] [AMDGPU][SPIRV] Correctly lower huge device function arguments (PR #176921)

[Steffen Larsen via cfe-commits]

steffenlarsen wrote:

Thanks, @mgcarrasco! 

> LGTM. Just asking to make sure that I got it right, the frontend now emits IR matching the allocation sizes. Before this PR, it wasn't due to the overflow. Some backends might not support the IR but that is a different issue, right?

Before these changes it would just keep the arguments by their direct type. For example, `deviceFuncWithHugeArg` in the new test would have `[6871947673600 x i64]` as its argument because the `classifyArgumentType` would think that it would fit into the remaining registers, due to the overflow making it look tiny.

https://github.com/llvm/llvm-project/pull/176921