Fix token counting, model resolution, and stabilize agent metrics integration tests. Update docker-compose and test runner with improved API key and model config.

fix: stabilize integration tests and improve agent drilldown UI compatibility

Refactor configuration flow (Day 0 vs Day 1 separation) and add startup scripts ...

- Stop AI provider settings from syncing back to config.yaml (stored in DB only now).
- Parameterize frontend port in docker-compose.
- Add start_server.sh and stop_server.sh scripts.
- Update .gitignore to exclude scratch files.

feat: Improve local testing and robustify agent loop ...

- Support native mode in run_integration_tests.sh

- Add error handling for ObjectDeletedError in agent_loop.py

- Use db_executor for heartbeat updates in node_registry.py

- Add new settings to config.py

- Add test for AuthService.handle_callback

- Update docker compose configs

feat: stabilize browser streaming, fix quality audit UI readability, and harden integration test harness

feat: enhance agent reasoning observability and remove turn duration clutter

feat(agents): full implementation of harness engineering and file sync fixes ...

- Implemented Area 1: Backend scaffolding, CRUD, schemas, and session locking
- Implemented Area 2: Agent execution loop, RAG integration, and Circuit Breaker
- Updated UI: Configure CRON triggers, Target Mesh nodes, and display system prompt in Agent Drilldown
- Fixed LLM fallback resolving vertex_ai_beta
- Fixed File System Sync pseudo-nodes (hub, local) raising 'Offline'
- Fixed FileSystemNavigator initialPath bug causing synced files not to render in UI Mesh Workspace

Fix gigabyte file sync race conditions and malicious path errors

Fix file sync integration tests and session auto-purge bug

test(integration): natively bypass Docker-in-Docker Dev Container constraints, fallback integration backend seamlessly via subprocesses

Refactor integration tests, introduce CUJ isolation, create E2E mesh runner script with no-rebuild, and clean up orphan files