Use OpenViking native search for research

[skyloevil]

What changed

• add an authenticated OpenViking HTTP client for search()
• constrain retrieval to L2 resources and hydrate each matched URI through /api/v1/content/read
• make ResearchAgent prefer OpenViking query planning and hierarchical vector retrieval
• retain JSONL and local-file fallbacks when remote search is unavailable
• exclude generated .abstract.md and .overview.md files from the local article fallback
• add operator-safe endpoint, identity, API-key environment, target URI, and timeout configuration
• document the native retrieval flow in English and Chinese

Why

The research path previously used local JSONL substring matching and filesystem recency, bypassing OpenViking's query planner, vector retrieval, and directory-tree traversal. Generated L0/L1 summary files could also be ingested as articles.

Impact

Research objectives now retrieve semantically relevant OpenViking L2 resources and pass full source text to downstream writing agents. Existing local/demo workflows remain compatible because remote search is opt-in and fallbacks are preserved.

Validation

• uv run --extra dev ruff check .
• uv run --extra dev pytest -q — 123 passed
• MockTransport coverage verifies authentication headers, request payload (level: 2), response parsing, body hydration, error handling, and fallback priority

A live search against the configured local server was not run because it would send the test query and retrieval context to the externally configured VLM provider.

[gemini-code-assist]

Caution

The consumer version of Gemini Code Assist on GitHub has been sunset. All code review activity has officially ceased.

[gemini-code-assist]

Caution

The consumer version of Gemini Code Assist on GitHub has been sunset. All code review activity has officially ceased.