JIT: Avoid remorphing general gtFoldExpr return

[jakobbotsch]

Potential alternative to #110516

[dotnet-policy-service]

Tagging subscribers to this area: @JulieLeeMSFT, @jakobbotsch
See info in area-owners.md if you want to be subscribed.

[jakobbotsch]

Looks like there are some x86 pre-order calls to gtFoldExpr that are problematic with this attempt.

[AndyAyersMS]

I am tempted to try something like the following. We remove all setting / checking for the morphed flag outside of morph.

Then we introduce an ambient state "allow remorph" which morph sets around remorphing the results of postorder calls to utilities like gtFoldExpr that make changes (we may also need more reliable means to tell if these utilities indeed made any changes). When this state is set morph doesn't assert if nodes it processes are already morphed. We can decide if we want to operationalize this and actually allow morph to skip remorphing of already morphed trees or just live with some redundant morphing.

Post global morph, we then verify everything got morphed at least once (eg has the flag set).

Or we change the flag into a count and assert post global morph that the count values are always at least 1 and not too large (so we're not doing crazy amounts of possibly redundant morphing).