feat: add optional you.com search integration

[mouse-value-add]

What this adds

An opt-in You.com web-search gather lane for search_structured. DeepGit's keyword and star-sorted topic queries both hit the GitHub search index, so they share a blind spot: repos that keyword search buries, or whose canonical descriptions omit the terms a developer would actually use (the same gap the topic sweep and the sparse-result retry already work around). A general web search sees those repos where they actually live in discourse — awesome lists, "X vs Y" comparisons, blog posts.

When enabled, the lane:

1. Queries the You.com Search API (0 LLM tokens) with the intent plus one keyword query
2. Extracts owner/name slugs from github.com/... links in result URLs, titles, descriptions, and snippets (site pages like github.com/about are filtered out)
3. Hydrates those slugs through the existing GitHubGraphQL client (repo:<owner>/<name> qualifier) so every web-surfaced candidate carries the same metadata + README + root tree evidence as the other angles
4. Folds the new candidates into the pool before the zero-token prefilter — the judge and adaptive controller are untouched

Opt-in by default-off

• No YDC_API_KEY → the lane never runs; behavior is byte-identical to today
• DEEPGIT_YOUCOM_SEARCH=0 → hard opt-out even with a key
• DEEPGIT_YOUCOM_MAX_QUERIES → tune web queries per search (default 2)
• HTTP errors / timeouts / no slugs found → returns no candidates and the pipeline continues on GitHub-only results (same failure contract as the optional semantic-recall layer)
• check_setup() now reports a youcom_web_search block so users can confirm the lane state

This follows the repo's existing pattern for optional capability: the [semantic] extra and enable_code_quality are both opt-in, and this lane adds no new dependency (httpx is already used for the GitHub client).

What changed

File	Change
src/deepgit/search/youdotcom.py (new)	Search API client + slug extraction + GraphQL hydration
src/deepgit/search/__init__.py (new)	Package exports
src/deepgit/pipeline.py	Lane wired in at step 2a (between gather and semantic layer), fully gated
src/deepgit/config.py	3 settings: youcom_api_key, youcom_search, youcom_max_queries
src/deepgit/mcp_server.py	check_setup() reports the lane
tests/test_youdotcom.py (new)	16 offline unit tests (extraction, gating, error paths)
README.md, .env.example, pyproject.toml	Docs + testpaths

Setup

export YDC_API_KEY=***   # https://you.com/platform/api-keys
# optional: DEEPGIT_YOUCOM_SEARCH=0, DEEPGIT_YOUCOM_MAX_QUERIES=2

Validation

• python -m pytest tests/ → 16 passed (extraction, gating, silent-failure, max-slug cap)
• ruff check src/deepgit/search/ tests/ src/deepgit/pipeline.py → clean; ruff format applied
• mypy src/deepgit/search/youdotcom.py → clean (config/pipeline have pre-existing errors on main, unchanged by this PR)
• check_setup() verified in all three states (no key / key / opt-out)
• Live check against You.com's keyless search surface confirmed the results.web → slug-extraction chain returns real repo slugs from real search results
• No key present: youcom_enabled() is False and gather_web_records is never invoked (asserted in tests)

Happy to adjust the shape if you'd rather have this behind a [youcom] extra, a different settings name, or as a gather-layer plugin point instead of inline in search_structured.