fix: Token counting now matches Claude Code UI context command

- Fixed token counting to include compact summary + user messages + assistant usage
- Added count_session_tokens() for consistent API interface
- Fixed isCompactSummary field name (was is_compact_summary)
- Includes cache_read_input_tokens for accurate counts
- Achieves 96% accuracy vs UI (6,341 vs 6,600 tokens)
- Closes API inconsistency issue for session token counting

🤖 Generated with Claude Code (https://claude.ai/code)

Co-Authored-By: Claude <[email protected]>

release: v2.0.1 - Fix Discord stop hook JSON structure bug

## Fixed
- get_latest_claude_message() now returns plain text instead of JSON structure
- Updated message.utils.get_text() to parse JSON in message.content field
- Discord stop hook now shows clean text instead of [{"type":"text","text":"..."}]

## Added
- Black box test with real JSONL data for stop hook API
- Towncrier changelog management setup
- Memory map update with Claude Code SDK patterns

## Changed
- Version bumped to 2.0.1 in pyproject.toml and __init__.py

🤖 Generated with [Claude Code](https://claude.ai/code)

Co-Authored-By: Claude <[email protected]>

Release v2.0.0 - Complete architecture redesign with Git-like disaste…

…r recovery

refactor: Add test factory for tool responses (Fix 3/10)

- Created HookTestFactory.create_tool_response_test()
- Reduces duplication by 90% (80 lines → 8 lines)
- Demonstrated in test_using_factory.py
- All tests pass (8/8)
- DRY violation fixed: 4x duplication eliminated

refactor: Replace magic strings in phase2 tests (Fix 2/10)

- Uses TestDefaults.SESSION_ID instead of 'abc123'
- Tests still pass (89/89 in phase2)
- No API changes
- Magic strings reduced: ~30 → 25

refactor: Add test constants module (Fix 1/10)

- Eliminates magic strings
- No existing code modified
- Zero breaking changes
- Violations: 18 → 18 (foundation only)