Core AI Overhaul: Schema, Tools, and Performance Gains

The schema management system was completely replaced by a declarative SchemaConvergenceService, which introspects the live database to ensure desired state.

This migration involved deleting the old SchemaService and run_pending_migrations, replacing them with logic that reads a declarative schema.sql.

The ACT orchestrator was replaced by a native tool-calling loop inside MessageProcessor, simplifying the system significantly.

Various memory and containerization issues were addressed by introducing CHALIE_SKIP_* environment variables and lazy loading mechanisms for models and caches.

Episodic relevance scoring was improved by replacing the broken clamp in query.py with sigmoid normalization to better differentiate relevance thresholds.

Numerous stale issues were closed, including those referencing deleted systems like uncertainty and ACT orchestrator.

Model loading was optimized by implementing ONNX pre-optimized loading across embedding and inference services.

Test suites were upgraded to use real database flows instead of hollow mocks for more reliable verification of decay and memory management.

• Replaced SchemaService with declarative SchemaConvergenceService.

• Replaced ACT orchestrator with native tool-calling loop in MessageProcessor.

• Introduced CHALIE_SKIP_* flags for memory-constrained test containers.

• Improved relevance scoring using sigmoid normalization for episodic relevance.

• Implemented ONNX pre-optimized model loading across services.

• Upgraded tests to use real flows instead of hollow mocks.