[clang] [llvm] [AMDGPU] Change CF intrinsics lowering to reconverge on predecessors. (PR #92809)

[Gang Chen via cfe-commits]

cmc-rep wrote:

@alex-t Just curious about the status of this PR. Both this PR and the register-allocation PR by CD will have significant impact to the generated code.  If we decide this is the right direction, then I feel it would be better to get it in earlier, so we can access its impact in our downstream work. 

https://github.com/llvm/llvm-project/pull/92809