Lint avoids revisiting Inlined.call

[som-snytt]

Fixes #24266

Restores empty check and "already seen" check for Inlined.call sites.

[som-snytt]

Would be nice to craft a test. One of the i15503 tests reports that it skips one tree.

llm4s reports locally:

sbt:llm4s> compile
[info] compiling 6 Scala sources to /home/amarki/projects/donderom-llm4s/target/scala-3.8.0-RC1-bin-SNAPSHOT/classes ..
SKIPPED 5874 of all 3572
SKIPPED 7 of all 49
[warn] -- [E198] Unused Symbol Warning: /home/amarki/projects/donderom-llm4s/src/main/scala/com/donderom/llm4s/Llm.scala:4:30
[warn] 4 |import fr.hammons.slinc.types.SizeT
[warn]   |                              ^^^^^
[warn]   |                              unused import
SKIPPED 0 of all 0
SKIPPED 8 of all 204
[warn] -- [E198] Unused Symbol Warning: /home/amarki/projects/donderom-llm4s/src/main/scala/com/donderom/llm4s/SlincLlm.scala:266:20
[warn] 266 |        for (batch, n) <- batches.zipWithIndex do
[warn]     |                    ^
[warn]     |                    unused pattern variable
SKIPPED 0 of all 0
SKIPPED 0 of all 0
[warn] two warnings found
[success] Total time: 7 s, completed Oct 27, 2025, 9:03:10 AM

where the counts are skipped call trees and the total (unique) call trees.

The debug message was emitted at reporting, i.e., after inlining, for various incremental runs. The first number is obviously significant. I did not investigate which trees are inlined many times. Note that -Vprint:inlining -Yplain-printer is unable to run to completion on llm4s. It would be nice to have -Vbrowse as in Scala 2. That is for another free weekend.

Without the fix, the build runs more than ten minutes, possibly much more than the limit of my patience.

[som-snytt]

The comment notes that these checks were tried in previous versions of the lint, but the glass of my knowledge about inlining is still only half full. Probably it was just wrong. Linting inlined code was removed in March for 3.7.0. (The wrong code was partially reverted and corrected at 0b3e859. I don't see a commit that tries to avoid visiting a call twice, but my vague memory is that I thought it was not effective. Obviously I lacked an effective test.)

[tgodzik]

LGTM!