[`eradicate`] Fix `ERA001`/`RUF100` conflict when `noqa` is on commented-out code (#25414)

## Summary

Fixes #25386.

For commented-out code with a trailing `# noqa: ERA001`, Ruff could
report both
`ERA001` on one line and `RUF100` (unused `ERA001`) on the next — which
is inconsistent.

`ERA001` uses `comment_contains_code()` to decide if a comment is code.
That function
was analyzing the full comment body, including trailing `# noqa:
ERA001`. The allowlist
treats `noqa` as non-code, so lines like `# ) # noqa: ERA001` were
skipped by `ERA001`
while a nearby line without `noqa` still triggered `ERA001`. That made
the `noqa`
look unused to `RUF100`.

The fix strips a trailing inline `# noqa...` before ERA001’s code
detection, so the
comment is evaluated on its code portion only (e.g. `)`). Suppression
and detection
then agree: `noqa: ERA001` is used when it suppresses a real `ERA001`
match.

the repro now shows ERA001 diagnostics without incorrectly flagging the
paired noqa: ERA001 as unused.

## Test Plan

- Added unit tests in `detection.rs` for `# ) # noqa: ERA001` and `# \t(
# noqa: ERA001`
- Extended `RUF100_5.py` with regression cases from #25386
- Ran `cargo test -p ruff_linter ruf100_5` (snapshot updated)
- Ran `cargo test -p ruff_linter comment_contains_code_with_multiline`
- Ran `uvx prek run --files` on the three changed files