Improve reliability and usage automation for v1.2.0

[aoright]

What changed

• recover stale Colima/Docker states automatically and deduplicate watchdog errors
• install Antigravity global instructions in the official ~/.gemini/GEMINI.md location while safely migrating older managed rules
• require automatic project context retrieval for project work and audit MCP client/tool usage without recording arguments or source content
• avoid repeated full walks with content-sensitive fingerprints for Git repositories and directories containing multiple Git repositories, while retaining a daily verification scan
• add compact critical backups that preserve project identity, collections, memories, history, embeddings, and audit events without copying rebuildable file indexes or web caches
• update macOS and Windows schedulers, documentation, release version, and manifest for v1.2.0

Why

The service watchdog treated a running Colima VM as healthy even when Docker was unavailable, Antigravity rules were written to a non-global path, unchanged projects were traversed every 30 minutes, and daily backups duplicated the full multi-gigabyte search index. These issues made the system less automatic and more resource-intensive than intended.

Impact

Project context retrieval becomes automatic in new Codex and Antigravity tasks. Service recovery is self-healing, indexing avoids unnecessary work, and daily backups are much smaller while full manual snapshots remain available.

Validation

• python3 -m unittest discover -s app/tests -v — 51 tests passed
• shasum -a 256 -c MANIFEST.sha256 — all files passed
• production service recovery verified with healthy Onyx and SearXNG containers
• production critical backup completed with SQLite integrity check ok
• 58-project indexing baseline completed with zero failures
• 132,138-document / 44-repository collection fingerprint measured at approximately 4.7 seconds on the development machine