test: coverage sprint — 65 new tests across 16 files

[leocamello]

Summary

Add 65 new unit tests across all 16 files with below-80% coverage to reach the 90% overall coverage target.

Test Breakdown

Category	Files	New Tests
Agent streaming	anthropic, generic, google, lmstudio, ollama, openai	12
Agent embeddings	ollama, openai	4
Agent quality	quality.rs	2
API handlers	completions.rs	10
API embeddings	embeddings.rs	5
CLI backends	backends.rs	4
CLI serve	serve.rs	6
Dashboard	handler.rs, websocket.rs	7
Discovery	mod.rs	8
Metrics	handler.rs	3
Queue	mod.rs	2
Total	16 files	65 tests

Coverage Areas

• Streaming: All 6 agent types now have chat_completion_stream tests (success + error)
• Embeddings: Ollama + OpenAI agent embeddings() method coverage
• API handlers: Priority extraction, budget header injection, error chunks, request recording
• CLI: Backend type auto-detection, config override propagation
• Dashboard: HTTP handlers, WebSocket update creation
• Discovery: IP selection, URL building, instance name extraction
• Metrics: Handler endpoints, budget stats computation
• Queue: Config access, priority dequeue ordering

Test Results

• All 878 tests pass (up from 813)
• Clippy clean (-D warnings)
• Formatting verified