Fixes involving SAM types

[odersky]

First, move ExpandSAMs to a group before HoistSuperArgs

Fixes #9391

This also uncovered two follow-on errors which were previously not detected, and which are fixed in the 2nd and 3rd commit.

[odersky]

It was instructive to see how this could be fixed:

#9391 looked at first mysterious, with a key-not-found crash in a tricky part of LambdaLift. But turning on -Ycheck:all revealed that there was a wrong owner after RefChecks. The phase group before RefChecks contained both the miniphase ByNameClosures for transforming by-name parameters and the miniphase ExpandSAMs for lifting closures into PartialFunctions. Since it was known that the example failed only if the two features were combined it was natural to try to separate the two miniphases. I split the phase group in the middle and that fixed the problem.

Since we do not want to create new phase groups if not absolutely necessary, I then moved ExpandSAMs to the previous phase group. That exposed two other errors, which in the end both had to do with the fact that methods created in the same group as RefChecks are not necessarily checked by it (the thinking is that these methods are compiler-generated anyway, so should be OK to start with). Now that ExpandSAMs ran in a group before RefChecks, some errors in the code it generated were reported.
So the larger part of the fix was to address these errors.

Morale: Kudos to our internal architecture and checking framework that allows us to move so fast to diagnose and fix bugs.