Undo struct promotion on "RetInd" code path

[kunalspathak]

When we undo the struct promotion, we revert all the promoted fields except in the code path where we would return the promoted field. We were keeping the promoted field leading to bad codegen.

No asmdiff on libraries/benchmarks/coreclr_test.

Since the original repro needed jitstress switch, I verified that this PR fixes the code we geenrate.

Fixes: #57912

[ghost]

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

Issue Details

---

When we undo the struct promotion, we revert all the promoted fields except in the code path where we would return the promoted field. We were keeping the promoted field leading to bad codegen.

Fixes: #57912

Author:	kunalspathak
Assignees:	-
Labels:	area-CodeGen-coreclr
Milestone:	-

[kunalspathak]

@dotnet/jit-contrib

[AndyAyersMS]

Changes look good. I assume fgMorphRetInd is relatively new?

Is there any way to add checking for cases where we mess up this undo?

[kunalspathak]

Changes look good. I assume fgMorphRetInd is relatively new?

It was added last year in #37745

Is there any way to add checking for cases where we mess up this undo?

I tried looking around and I didn't see a good place to have this check. What I ended up doing is introducing a field lvUndoneStructPromotion and then, during mark local reference, whenever we see that we are increasing the reference count of a struct field, we make sure that its parent struct didn't have lvUndoneStructPromotion set.

By running superpmi replay on coreclr, it hits the newly added assert on exactly same tests.

[kunalspathak]

/backport to release/6.0

[github-actions]

Started backporting to release/6.0: https://github.com/dotnet/runtime/actions/runs/1196839095