Revert "Approximate MatchTypes with lub of case bodies, if non-recursive" in 3.5.0

[WojciechMazur]

Reverts #19761 in 3.5.0-RC6 as discussed in #21258 and accepted by the core compiler team.

[dwijnand]

Why is this reverted in 3.5.0?

[WojciechMazur]

Conclusion from last core meeting:

#19761 will be reverted on every release branch and on the main. Then, we will release a hotfix for the 3.4 line. We will decide what to do with that change before 3.6. We will set up automatic tests to find similar compatibility problems in the future.

[sjrd]

In order to elaborate on the motivation for that decision. What came out of discussions is:

• The new state after the change looks better per se. We probably want it.
• We definitely revert it in 3.4.x.
• We don't want to ship it in 3.5.0 "just because it's there already". We want to do so very intentionally.
• We don't really have time to do it properly (intentionally, with good accompanying doc, perhaps with migration flags/warnings, etc.) for 3.5.0.
• Therefore we revert in 3.5.x as well, and take the time to do it properly for 3.6.0.

[dwijnand]

Seems like a standard that I don't feel has been equally applied for other improvements, except for the implicit prioritisation changes and change fixes. I wish I could have participated in this decision. But thank you for explaining the motivation.