-Xsource:3 now respects refinements by whitebox macro overrides

[som-snytt]

Fixes scala/bug#12645

[lrytz]

This fix seems a little arbitrary to me.

The way I understand whitebox macros is that the type of a macro invocation is the inferred type of the expansion. But this PR is changing the rules of type inference while type checking a macro expansion, which is something else.

On the other hand, since this is about macros which anyway don't cross-build with Scala 3, I guess it's benign. The underlying complication is the staging aspect, macro expansions are compiled with the user's compiler flags, and the quill macros require -Xsource:3 to be disabled to work as intended.

[som-snytt]

Edit: actually, maybe the only relevant part of my comment (I haven't had coffee yet) is

Possibly, the macro should be required to generate the signature if it wants it.

I think this is why whitebox magic such as https://docs.scala-lang.org/overviews/macros/typeproviders.html works.

I agree that once you're inside the expansion, it seems to weaken language guarantees.

The quill code emits def quoted = ast; def ast = $etwas; this is the required member refinement that is inferred. So it's turtles all the way down.

Possibly, the macro should be required to generate the signature if it wants it.

I couldn't find (quickly) a way to test if the current macro is whitebox, so it merely checks if any macros are getting expanded.

As a footnote, there is still a bit of old code in assignTypeToTree for macros that I think isn't used. Also noting that Scala 2 macros are still experimental.

[lrytz]

Possibly, the macro should be required to generate the signature if it wants it.

I also think this would be the right thing to do. But more pragmatically, I think we should just merge this PR. As I said before, it seems benign, and it avoids causing problems for people that don't even know they are relying on our experiments.