Refuse variable inlining across lazy type alias boundaries

[yangfan-yf-yf]

Inlining a variable used by a PEP 695 type alias can change when its initializer runs and which binding the alias resolves. For example, inlining target below changes the output from True to False:

original = int
target = original
type Alias = target
original = str
print(Alias.__value__ is int)

I added a preflight check that raises RefactoringError when a variable reference occurs within a type alias in the selected refactoring resources. This preserves the source instead of moving the initializer into a lazy context. A partial inline of an eager reference remains available with only_current=True, remove=False; removing a binding still used by a type alias in those resources is refused.

The regression tests execute source before and after the operation, covering side effects and late name binding as well as cross-module references, type parameter bounds, multiline and Unicode aliases, partial inlining, and unrelated/shadowed-name controls. On the original unmodified 2bd17a8 base (tested on 2026-09-29), 13 of the 17 new cases fail; all 17 pass with this change.

This follows the runtime-semantics testing direction in Discussion #875. It addresses ordinary variables referenced inside type aliases, separately from #874's handling of selecting a type-alias definition for inlining. Deferred function/variable annotations and type parameters outside type aliases are outside this change.

Validation after updating against master 22f5e50:

• Windows, Python 3.12.3: full suite, 2166 passed, 12 skipped, 5 xfailed.
• Windows, Python 3.14.7: new regression suite plus existing inline suite, 119 passed.
• All five configured pre-commit checks passed for the changed files.
• The type-alias guard remains unchanged; code samples follow the current textwrap.dedent() test guideline.

Checklist:

• Added regression tests.
• Updated CHANGELOG.md.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 95.36%. Comparing base (22f5e50) to head (5232586).

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #890      +/-   ##
==========================================
+ Coverage   95.34%   95.36%   +0.02%     
==========================================
  Files         134      135       +1     
  Lines       26762    26885     +123     
==========================================
+ Hits        25516    25639     +123     
  Misses       1246     1246              

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

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