Respect the magic trailing comma in starred type parameter lists

[dchaudhari7177]

Summary

Fixes #5243. A magic trailing comma explodes a PEP 695 type-parameter list of plain type vars, but was silently ignored once the list contained a *TypeVarTuple or **ParamSpec:

class C[
    T,
    *Ts,
    **P,
]:
    pass

collapsed back to

class C[
    T, *Ts, **P,
]:
    pass

Root cause

is_split_before_delimiter / is_split_after_delimiter deliberately skip a star that is_vararg() recognises, with the comment:

# * and ** might also be MATH_OPERATORS but in this case they are not.
# Don't treat them as a delimiter.

is_vararg() decides that by checking the star's parent against VARARGS_PARENTS, which listed the argument/lambda nodes but not the PEP 695 typevartuple and paramspec nodes:

leaf='*'  parent_sym=typevartuple   is_vararg=False
leaf='**' parent_sym=paramspec      is_vararg=False

So the stars in a type-parameter list were scored as * and ** math operators, and that delimiter outranked the comma when the line was split — which is why the trailing comma stopped mattering.

Adding both nodes to VARARGS_PARENTS restores the intended "not a delimiter" treatment.

Scope

Fixes def, async def, class and type aliases alike, matching the issue. The non-magic case is unchanged — def all_in[T: int, U: (bytes, str), *Ts, **P]() still stays on one line (pinned by the existing fixture case).

Tests

Added magic_starred (def), MagicStarred (class) and MagicStarredAlias (type alias) to tests/data/cases/type_params.py, right beside the existing plain-type-var magic[Trailing, Comma,] case.

Reverting only src/ fails the fixture with exactly the reported collapse, on all three constructs:

-    *Ts,
+    T, *Ts, **P,
-    *Ts,
+    T, *Ts, **P,
-    *Ts,
+    T, *Ts, **P,

Verification

• pytest tests/ -k "not symlink" → 464 passed, 3 skipped, 8 subtests passed — no formatting regressions across the fixture corpus
• The 6 deselected symlink tests fail on a clean tree on my machine (OSError: WinError 1314 A required privilege is not held by the client — Windows needs elevation to create symlinks), verified by stashing; unrelated to this change

I'll add the CHANGES.md entry under Stable style in a follow-up commit on this branch, since the file asks for the PR number rather than the issue number and I only get that once this is opened.

[github-actions]

diff-shades results comparing this PR (94cfff8) to main (51abf53):

--preview style: no changes

--stable style: no changes

---

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

[cobaltt7]

Thanks!