[JumpThread] Fix `JumpThreading` pass to skip merging when both blocks contain convergence loop/entry intrinsics. (#170247)
Fixes: https://github.com/llvm/llvm-project/issues/165642. After this
fix, optimization passes for the example in the bug.
[LLVM
Spec](https://llvm.org/docs/ConvergentOperations.html#llvm-experimental-convergence-loop)
states that only a single loop / entry convergence token can be included
in a basic block.
This PR fixes the issue in `JumpThreading` pass so that when a basic
block and its predecessor both contain such convergence intrinsics, it
skips merging the two blocks.