Change scope nested CFG passes to use a stack instead of recursion

[jenatali]

Fixes #8224

[github-actions]

✅ With the latest revision this PR passed the C/C++ code formatter.

[llvm-beanz]

This does not conform to our coding standards some obvious violations include:

• Do not use using namespace std
• Pick the right data structures

[llvm-beanz]

Thank you for the cleanups! I just have two small nits, otherwise this looks good to me.

[bob80905]

I can't wrap my mind around how this is logically equivalent to the previous code, this looks like a behavioral change. Lines 1054-1074 here are the only lines I can't see the equivalent iterative translation for. Can you explain why we continue here? Maybe that would help explain the previous lines to me also. Other than those lines, this change looks good to me.

[jenatali]

The previous logic would:

• Check for work to do for the target block
• For each successor:
  • Recurse
  • After recursion for a specific block, update the reachable blocks
• After all recursion, update the late watermark and append to BTO

The new logic will:

• In an iteration, check if this is the "before recursion" phase (based on the initialized flag), and if so, do the pre-recursion work.
• Process one successor: For the current successor, if we haven't processed it, then push a stack frame, break out of the successor loop, and immediately proceed back to the top of the iteration loop (i.e. begin processing that new frame). If we've already processed the successor, then iterate until we find one we haven't processed, or until we've checked all of them.
• If we didn't push a recursive stack frame, then process the post-recursion work.

[bob80905]

The continue makes sense now. There is an extra DXAssert that is being introduced that I don't think was in the previous version, and it might be redundant since this assert already exists earlier in the function on what appears to be an already-processed frame. But otherwise LGTM.