Make versioner skip loops that have become acyclic during the pass

[jdmpapin]

The region structure is expected to be a loop. If it isn't, then versioner may misbehave, e.g. it may crash. And besides, there's no point in trying to version an acyclic region anyway.

HCR guard versioning removes edges from the CFG immediately (unlike versioning of regular conditionals/virtual guards, which leaves a trivially foldable conditional tree in place). The removal of particular edges can change loops into acyclic regions, so skip any region that is no longer a loop.

(OSR guard versioning also removes edges immediately, but the targets of those edges are not within any loop, so their removal can't change loops into acyclic regions.)

Fixes eclipse-openj9/openj9#15061

[vijaysun-omr]

Jenkins test sanity.functional,extended.functional all jdk17

[vijaysun-omr]

Do we think we should consider not allowing HCR guard versioning to change trees immediately (thus creating this problem) in the future ?

[jdmpapin]

Possibly. When this does happen, it's better to avoid versioning the "loop" (as done in this commit), but it should be an extremely rare case, so I don't expect any performance benefit

To understand how rare: Reproducing this reliably (since the repro attached to eclipse-openj9/openj9#15061 was intermittent), I had to create a truly bizarre configuration of two nested loops:

• The outer loop had to contain an HCR guard that exits the loop on the passing edge.
• The rest of the outer loop's body still had to be reachable (in the CFG) after deleting the HCR guard
• Versioner's HCR guard search had to be able to tell that the rest of the outer loop's body (including the inner loop) was unreachable, so it had to be initially reachable only via the HCR guard's failing edge or via another conditional which gets versioned.
• The inner loop had to have its back-edge dominated by another HCR guard's failing edge.

[jdmpapin]

Jenkins build all