Preserve inline comments inside annotation subscripts

[Pedro-Muller29]

Description

Fixes #4733.

Black crashes with INTERNAL ERROR: produced different code on the second pass of the formatter on inputs like:

aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa: list[  # pyright: ignore[...]
    int
] = []

Root cause

The collapsed line is one char over the limit thanks to the inline comment
on the subscript's [. That's enough for run_transformer to try the
FORCE_OPTIONAL_PARENTHESES second opinion, which wraps the annotation in
invisible parens and pushes the comment out of the subscript. On the next
parse the comment lands on a different leaf, the splitter picks a different
shape, and the second-pass stability check blows up.

Fix

Skip the second opinion in run_transformer when result[0] only goes over
because of comments attached to a subscript [ (LSQB). The comment stops
migrating, formatting converges in one or two passes, and it stays where
the user wrote it.

Checklist - did you ...

• Implement any code style changes under the --preview style, following the stability policy?
• Add an entry in CHANGES.md if necessary?
• Add / update tests if necessary?
• Add new / update outdated documentation?

[github-actions]

diff-shades results comparing this PR (24ac9a0) to main (9cbd95f):

--preview style: no changes

--stable style: no changes

---

What is this? | Workflow run | diff-shades documentation

[cobaltt7]

With this change, the original test case is reformatted badly:

 def foo():
     possibly_redundant_lowlevel_checkpoints: list[  # pyright: ignore[reportUnknownVariableType]
         cst.BaseExpression
-    ] = field(default_factory=list)
+    ] = field(
+        default_factory=list
+    )

To be fair this is an improvement, but I think there's more work to be done on this issue.

[Pedro-Muller29]

With this change, the original test case is reformatted badly:

 def foo():
     possibly_redundant_lowlevel_checkpoints: list[  # pyright: ignore[reportUnknownVariableType]
         cst.BaseExpression
-    ] = field(default_factory=list)
+    ] = field(
+        default_factory=list
+    )

To be fair this is an improvement, but I think there's more work to be done on this issue.

Both addressed in 5e02bac (moved tests under tests/data/cases/) and 24ac9a0 (keeps the omit-based RHS split so field(default_factory=list) no longer explodes).

CI is red but seems unrelated to my changes. Looks like Click 8.4.0 broke the FakeParameter stub at tests/test_black.py:108.