Teach the fix-error workflow to read error construction code before
proposing fixes. Instead of hardcoding knowledge about specific error
types (e.g., listener leak categories), the AI is instructed to:
1. Search for where the error is constructed in the codebase
2. Read the surrounding code to understand conditions, categories, thresholds
3. Use that understanding to determine the correct fix strategy
Includes a listener leak example showing how reading ListenerLeakError
construction in event.ts reveals the dominated/popular classification.
Relates to #289777
