[llvm] [AMDGPU] Add intrinsic readanylane (PR #115696)

[Xin Russell Liu via llvm-commits]

GinShio wrote:

I thought maybe we should consider `uniform when defined` or `uniform when used`.

If it's former, this intrinsic should be to undefined behavior when defined is divergent. That means, this example is invalid, user shouldn't generate `readanylane` in this case.
If it's latter, keep the attribute `convergent` and modify the change code. Like Joy said, just allow suck it.

Looks like, the former has more opportunities to optimize.

https://github.com/llvm/llvm-project/pull/115696