Remove ^ in capture set syntax

[noti0na1]

This PR simplifies the capture set syntax by removing unnecessary ^ symbols in capture sets.
This change is the first step making the syntax cleaner and more consistent for capture polymorphism.

1. Added a new InCaptureSet mode in Mode.scala to handle capture set annotation typing;
2. Modified the parser to remove the ^ suffix from capture set syntax
3. If InCaptureSet is set and a term ident or select is not found, retry typing as a type wrapped in CapsOf.

[bracevac]

Coolio!

[bracevac]

Let's rebase this on #22539 and wait until after the release cutoff before merging.

[noti0na1]

I don't think we need to rebase onto the separation checking PR. The change here is only inside the Typer and should be independent to the core cc phase.

[odersky]

How about all the test cases that need to be updated? I think it's better to rebase.

[odersky]

This scheme has the advantage that the code is very simple. But there is an issue: Assume we have

object C
...
def foo[C^](x: A^{C}) = ...

then A^{C} will refer to the outer C and probably fail type checking. What's worse, if the outer C actually reads

val C: Capability = ...

then we get scope inversion. The outer C will be selected over the inner C. I think we should try to avoid these problems by using a scheme similar to CBCompanion where when we run the Namer on [C^] we add C under a term name and as a type name in the same scope. Then we will always find the innermost C.

[noti0na1]

I see, if we keep searching terms first, we need to create a "fake" symbol for every type bounded by CapSet to avoid scope inversion.

[noti0na1]

I need to wait for the syntax change in #22758, so it is easier to tell which type def is a capture variable.

[odersky]

I agree. I first thought this was independent of the syntax changes, but maybe that's not the case.

[bracevac]

@noti0na1 it's fairly complete now, modulo adapting the test cases.