Investigate whether crossgen composite mode inlining is behaving as intended on all platforms #118028
Description
#109315 and #117823 and others appear to be situations where a method with AggressiveInlining was not inlined as expected. These failures appear to only occur when crossgen/r2r is operating in composite mode, and they are platform specific - I am unable to reproduce them on Windows and they tend not to happen on Windows-based CI either.
It's likely that our inlining behavior here changed sometime in the past few releases and the change wasn't noticed due to the tests being confined to the outerloop. We should make certain that the new behavior is what we want, and figure out why it doesn't consistently reproduce on Windows, because having inlining behavior depend on OS is counter-intuitive.
So far none of the test failures indicate an actual regression, however, just a behavior change.