fix: Remove "One of these files" error

[eed3si9n]

Problem

Fixes #21973
Fixes #17394
Ref sbt/sbt#7726
Ref #7652

Currently the ambiguous symbol detection raises false positives when used together with -Wunused:imports. sbt 2.x plugins are running into this bug.

Solution

Remove the faulty ambiguous symbol detection for now.

[KacperFKorban]

I think that we should fix the implementation of computeMembersNamed instead. It looks like all those issues are caused by some bug in that function.

[eed3si9n]

#17394 has been open since May 2023, and I don't think anyone has sent a PR to fix it thus far. Wouldn't you agree that having no linter is better than a false-positive liner in a compiler for the current stable releases? (3.3.x, 3.5.x) While fixing forward would be ideal, every project has finite resource and priority list, so we should be ok with dropping bad linters.

This PR adds a test, so when the fix is implemented hopefully before the next minor series, it can come back as nightly first etc.

[dwijnand]

Wouldn't you agree that having no linter is better than a false-positive liner in a compiler for the current stable releases? (3.3.x, 3.5.x)

I'd agree with you, on that basis, if the change were -Wunused:imports specific. But it's not, it's general to all SymDenotation#dropStale.

[KacperFKorban]

I agree that it is not ideal to have an issue like this lay around for so long.
@Gedochao Should we assign someone to fix the underlying issue this cycle?

[eed3si9n]

@dwijnand When there's an interference between feature A and feature B, I'm suggesting that we drop (revert PR when possible) either A or B, not A ∩ B, since likely the bug can pop up with some other intersection as well. But I guess it's case by case, since we can't just drop macro and extension methods when they interfere.

[Gedochao]

I agree that it is not ideal to have an issue like this lay around for so long. @Gedochao Should we assign someone to fix the underlying issue this cycle?

Yup, we'll try to squeeze it into 3.6.4.