[ClosureLifetimeFixup] Cleanups get CleanupLocs.

[nate-chandler]

Otherwise at Onone we can get downstream debug info verification failures.

[nate-chandler]

@swift-ci please test

[nate-chandler]

@swift-ci please clean test macOS platform

[atrick]

huh, I bet this needs to be fixed all over the code. What actually changes with CleanupLocKind? Do certain kinds of instructions require it?

[nate-chandler]

@atrick Quite possibly. What changes is that when verifying debug scope holes, this condition is true

// lib/SIL/Verifier/SILVerifier.cpp:5813
if (SI.getLoc().getKind() == SILLocation::CleanupKind)
  continue;

so instructions that are created with autogenerated locs "<compiler-generated>":0:0 aren't required to participate in scope reentry expectations.

Luckily, that verification is only done at Onone, so many places wouldn't need to be updated.

[atrick]

so instructions that are created with autogenerated locs "<compiler-generated>":0:0 aren't required to participate in scope reentry expectations.

oh, I thought it was sufficient just to use an autogenerated location. That's why we're not doing the CleanupLoc think consistently