Keep overlong star imports stable with Black

[Spidy092]

Summary

• Keep over-long from ... import * and from ... cimport * statements intact when they cannot be safely wrapped.
• Preserve trailing comments and the existing NOQA wrapping mode behavior.
• Add regression coverage for direct wrapping and the Black profile.

Fixes #2649

Testing

• python3 -m pytest -q tests/unit/test_wrap.py (8 passed)
• ruff check isort/wrap.py tests/unit/test_wrap.py
• ruff format --check isort/wrap.py tests/unit/test_wrap.py
• Direct reproduction: output remains unchanged, compiles, and is idempotent under profile="black".
• The full tests/unit collection is blocked in the isolated harness because the test-only dependencies black, hypothesis, and colorama are unavailable.

[Manny7717]

Verified at head 9ef16a1 — this resolves the isort↔black non-convergence from #2649 exactly as reported.

Regression-proven: the 3 rewritten star-import spec tests FAIL on base 131f4ad (backslash output still emitted: import \\ + *) and PASS at head (statement returned untouched). 8/8 in tests/unit/test_wrap.py at head.

End-to-end convergence probe (real CLI, black 26.3.1, profile=black, over-long from ... import *):

• Base 131f4ad: isort backslash-wraps → black rejoins to one line → isort wraps again → alternates forever (no fixed point — reproduces #2649 verbatim, incl. "Fixing" on every isort pass).
• Head: isort leaves the line → black leaves it → isort leaves it. Stable after one pass.

No collateral: full tests/unit at head 623 passed / 2 failed / 1 skipped — the 2 failures are the pre-existing issue_909/issue_938 FileNotFoundError env-noise tests (identical on clean main across all my prior runs). The endswith("import *") condition is stricter than the old endswith("*") (only star from/cimport lines match; short star imports still return via the length check before this branch; comments preserved via the earlier # split). The #2267 corruption fix (never emit import (*)) remains satisfied — the statement is now kept whole and valid instead of backslash-split.

Non-blocking nit: the # skipcq: PY-R1000 directive comments a function whose complexity this PR actually reduces (branch removed, ~6 lines deleted); the two-line comment is also redundant. If the DeepSource gate still trips at the threshold, fine to keep, but consider dropping both if the removal alone clears it — external-service gates shouldn't pin unrelated comments in the tree.

[DanielNoord]

This is not using the black profile. Why are we changing behaviour here?

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 99.37%. Comparing base (131f4ad) to head (9ef16a1).
⚠️ Report is 15 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #2652      +/-   ##
==========================================
- Coverage   99.37%   99.37%   -0.01%     
==========================================
  Files          41       41              
  Lines        3181     3179       -2     
  Branches      686      686              
==========================================
- Hits         3161     3159       -2     
  Misses         12       12              
  Partials        8        8              

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[DanielNoord]

Superseded by #2665 which is a more targetted fix.