…objects

The regex coverage test was failing to track pattern matches because Prism creates new `RegExp objects` when adding flags (e.g., `g` or `d`) during tokenization. The original approach intercepted `exec()` on individual regex objects, which missed matches on the newly created `RegExp` instances.

Changes:
- Replace `String(regex)` key with normalized source+flags key to match patterns even when Prism creates new `RegExp` objects with different flags
- Switch from per-regex interception to global `RegExp.prototype.exec` interception to catch all pattern matches, including on new `RegExp` objects
- Use a simple loop instead of `String.replace` in flag normalization to avoid triggering our own `RegExp.exec` interception (which caused infinite loops)

This fixes the tracking issue for patterns like the markdown code block pattern that were previously reported as untested despite being used in tests.

Plus, add tests.