Feat/model library worker reconnect compose

[rwilliamspbg-ops]

Summary

This branch hardens the model library, worker lifecycle, GUI backend wiring, and Docker runtime setup for the Mohawk Inference Engine.

What Changed

Model library

• Added persistent local model library tracking in mohawk/models/loader.py.
• Enabled HuggingFace downloads to be registered automatically in the local model index.
• Added validation for blank model IDs and safer HuggingFace download argument handling.
• Added tests for local model registration, HuggingFace download registration, and empty-ID rejection.

Worker lifecycle and secure execution

• Added worker join, leave, and reconnect APIs.
• Added failover/reshare handling for worker churn.
• Fixed deterministic shared-key derivation for secure worker/controller communication.
• Improved slice serialization to preserve shapes and support backward compatibility.
• Added regression tests covering worker lifecycle and reconnect flows.

GUI backend wiring

• Rewired the desktop GUI to use real controller API endpoints for models, workers, queue, sessions, metrics, and security actions.
• Improved API error propagation so backend failures surface meaningful details in the UI.
• Added a controller service backend that exposes the GUI-facing API contract.

Docker and CI

• Corrected compose files and Dockerfiles so the runtime commands match the service ports.
• Added CI compose smoke validation.
• Fixed the controller Dockerfile healthcheck/exposed port mismatch so the image matches the actual uvicorn service port.

Validation

• Full test suite: 87 passed, 4 skipped
• Focused controller/model tests: passed
• Compose config validation: passed

Notes

• Docker compose runtime still requires a machine with a working Docker daemon for full up --build execution.