fix(session): tolerate invalid UTF-8 in saved sessions

[jipeng6036-del]

A saved session containing an invalid UTF-8 byte, or truncated in the middle of a multibyte character, raises UnicodeDecodeError before JSON parsing. This crashes --resume and /sessions, despite the existing handling for corrupt or truncated JSON.

Catch UnicodeDecodeError in both session readers: resume returns None, and listing skips the unreadable session while retaining valid sessions. The files are left untouched. Four regression cases cover invalid bytes and truncated UTF-8 in both paths, including a valid Chinese-language session alongside a corrupt file.

Validation on Windows, Python 3.13.12:

• Before the fix: all four new regression cases fail with UnicodeDecodeError.
• python -m pytest tests/ -q: 229 passed.
• ruff check corecoder tests, python -m compileall -q corecoder tests, python -m build, and python -m twine check dist/*: passed.

AI assistance: this change and its regression tests were prepared with OpenAI Codex.

[he-yufeng]

Thanks for the catch, @jipeng6036-del. Verified locally: a session file with invalid UTF-8 raises UnicodeDecodeError before json.loads, crashing --resume, and one such file kills the whole /sessions listing. Landed the same fix as my own commit in 1f201a2 with regression coverage for both readers, so closing this one. Credit for the report stands.