[llvm] x86: fix musttail sibcall miscompilation (PR #168956)

[Folkert de Vries via llvm-commits]

================

----------------
folkertdev wrote:

Right, the x86 backend notion of a sibcall is a little stronger currently than I think the original term intended. In the x86 backend it means that a sibling can be called without moving arguments around on the stack at all.

That is a restriction that could now be lifted, but it created a lot of test changes earlier, so I'd prefer to take this on in smaller steps.

https://github.com/llvm/llvm-project/pull/168956