[clang] [CUDA] Remove sema check of function declaration with variadic argument (PR #161350)

[Jon Chesterfield via cfe-commits]

JonChesterfield wrote:

There is no abi break here as far as I can tell.

The variadic lowering pass lays things out the same way cuda's printf works and matches the documentation on ptx. As in if you've written a variadic function in ptx, calling it from IR is expected to work.

Also, it's going from refuse to compile to does compile, so there's nominally no pre-existing code to be compatible with.

There may be bugs in the lowering pass. I really should revive the x86/aarch64 implementations and/or chase wasm to get it running on non-gpu targets.

https://github.com/llvm/llvm-project/pull/161350