Preserve reload worker failure exit status

[hansu650]

Fixes #269.

When a reloaded worker failed (for example, while importing an application with a new syntax error),
un() initially captured its non-zero exit code and removed the failed process. The final cleanup then called _join_exited() again on the now-empty process list and assigned its zero result back to �xitcode, so the CLI reported success.

This keeps the cleanup join but preserves the already-recorded failure code. The regression test models a clean initial worker reload followed by a replacement worker exiting with status 1.

Validation:

• Manual hypercorn --reload reproduction: syntax error changed the shell status from 0 before the fix to 1 after it
• pytest tests/test_run.py -q (1 passed)
• Black, isort, and flake8 checks pass
• mypy src/hypercorn/run.py tests/test_run.py passes
• sdist/wheel build and wine check pass

The full Windows test run reaches 190 passed with 7 unrelated baseline failures in Unix-only socket/path tests, e2e binding, and HTTP/2 CONNECT validation against the latest h2.