Critical fix: accept-process-output CPU spin bug

Fixed critical bug where daemon would spin at 100% CPU and become unresponsive.

**Root Cause:**
- Previous 'fix' changed (accept-process-output process 0.1 nil t) to (accept-process-output process 0.1 nil nil)
- The last argument 't' means 'don't block', 'nil' means 'DO block'
- This caused daemon to block in tight loop, consuming 100% CPU

**The Fix:**
- Restored 't' as last argument to prevent blocking
- Ensures non-blocking poll in while loop
- Daemon now runs at 0% CPU when idle

**Testing:**
- Daemon responsive at 0% CPU
- No blocking or spinning
- Timeout function works correctly

Release 2026.03.30.2 - Performance optimization and validation improv…

…ements

- Performance: Eliminate redundant score extraction in summarize-results
- Validation: Add commit hash validation in track-commit
- Both improvements passed grader with 9/9 scores

Release 2026.03.30.1 - Auto-sync main/staging branches fix

Release 2026.03.30 - Robust shell timeout and substring safety fixes

Auto-workflow production ready

- 50+ real code fixes from LLM agents
- Researcher switched to DashScope (faster)
- Metadata cache persistence
- Tree-sitter validation helper
- Line filtering helper
- Model-id caching optimizations
- Submodules updated (gptel, gptel-agent, ai-code)

Autonomous Research Agent: Complete implementation

Features:
- Decision logic: 70% grader + 30% code quality
- Subagents: analyzer, grader, comparator integrated
- LLM degradation detection
- Code quality scoring (docstring coverage)
- TSV logging with code_quality column

Pipeline:
worktree → analyzer → executor → grader → benchmark → code-quality → comparator → decide

Tests: 74/74 (grader + retry)
Commits: 37

Log timeout experiments to TSV for audit trail

- Timeout path now writes to results.tsv before cleanup
- Ensures overnight runs have complete experiment history

Add mock mode for testing auto-workflow callback chain

- Increase timeout to 900s for slow API responses
- Add gptel-auto-experiment-mock-mode flag
- Add gptel-auto-experiment-test-mock function
- Test callback chain without API calls

Add debug logging to auto-workflow callback chain

- Trace each step of the experiment workflow
- Log timeout, grade, benchmark, decide outcomes
- Helps diagnose async callback issues

Fix auto-workflow worktree creation and deletion

- Use absolute paths to prevent nested worktrees
- Fix branch name path construction
- Add git worktree prune after deletion
- Set default-directory for git operations