Find annotation args in inline expansion

[som-snytt]

Fixes #24894

Annotation args can be inline transparent, but Inlining phase doesn't inline into annotations.

The Inlined node must be handled when emitting the annotation and also when inspecting args.

[som-snytt]

I'm glad the checks are green, but I had a local failure that final val was no longer stable in a pattern. I tossed in some StableRealizable and UncheckedStableAnnot but apparently that is not necessary. Probably I had a stale binary in .ivy2.

I wonder if there's a way for the test rig to warn me if a failure includes a suspicious artifact.

At the moment:

[error] -- [E046] Cyclic Error: /home/amarki/projects/scala3/staging/test/scala/quoted/staging/repl/StagingScriptedReplTests.scala:11:39
[error] 11 |class StagingScriptedReplTests extends ReplTest(ReplTest.withStagingOptions) {
[error]    |                                       ^
[error]    |                             Cyclic reference involving val <import>

I'm not going to investigate, but it would be nice if the test suite just ran (locally). Elsewhere, a PR includes tweaks to help run under JDK 25. Spurious warnings and errors are demoralizing and distracting.

I'll add a simple test for the stability check in patterns. (Bootstrapping would fail, as it did for me spuriously, if that were broken. It complained about the vals in concurrent.duration which are aliases of Java enums.) I'll also ask myself why it doesn't need more help with stability. The check is checkLegalValue in typer.

[som-snytt]

Moving the second fix for final val enum to a separate PR, including @inelidable or @inelisible.

[jchyb]

Thank you for the fix! (and for splitting the PR up - from what I understand, the API changes for stdlib are effectively frozen until 3.10, so the PR would have had to wait until then).

Overall the fix looks great, but I do have one question about the allArguments and allTermArguments parts.

[jchyb]

I don't really understand this part. It seems like we are removing the Inlined node from the function of Apply, rather than the argument, which seems unrelated to the rest of the PR.
Like for annotation @annot([Inlined] "someString")) allArguments will still return [Inlined] "someString"

[som-snytt]

Likely! I didn't get back to write a test, a mental block. I'll follow up.

[som-snytt]

Dropped the ill-considered allTermArguments tweak and added a unit test for what is actually witnessed. Did not fiddle with adding a helper for convenience.

[jchyb]

Thank you!