gh-157660: Refresh cached TLBC arrays after growth

[liuzhijie-0614]

Fix persistent Invalid tlbc_index errors when an existing thread grows a code object's TLBC array without changing tlbc_generation (#157660).

When an index exceeds the cached capacity, refresh that entry once using direct remote reads, then retain the existing bounds checks and fallback. Ordinary cache loads continue to use paged reads.

The regression test uses RemoteUnwinder.get_stack_trace() with all worker threads created before sampling. It verifies recovery with the same unwinder, correct line numbers, and both frame-cache settings.

Validated on upstream main (e2ff498), Linux x86-64, with a free-threaded debug build:

• ./python -m test test_external_inspection --timeout=180: passed; 41 tests, 2 skipped.
• The original issue reproducer: no bounds or other errors before growth, after growth, or with a fresh unwinder.

• Issue: _remote_debugging retains stale TLBC arrays after growth in free-threaded builds #157660

[python-cla-bot]

All commit authors signed the Contributor License Agreement.

[liuzhijie-0614]

This is ready for review. The TLBC regression test passes locally, including repeated runs with frame caching enabled and disabled. The two failing CI jobs appear unrelated to this change, as noted above.