[llvm] [Verifier] Reject va_start in non-variadic function (PR #88809)

[Jon Chesterfield via llvm-commits]

JonChesterfield wrote:

The best case interpretation for the construct is UB - we could do something like replace it with undef if it occurs, but as the construct is nonsense and clang rejects it in sema anyway, I don't think this arises in reality. I'd like it to be an IR invariant so I don't have to handle it in varargs lowering.

https://github.com/llvm/llvm-project/pull/88809