Scout Docs Versioning Release Preparation

Scout now has a doc_root setting and a new s3 release that reads pages from docs/v2. The s3 eval scores 15/18, because docs/v2 doesn't exist yet. s2 still scores 18/18. I committed as c42532d, with only scout.py and releases/s3.yaml.

• scout.py: Release has a doc_root field, default docs, relative to agents/scout. read_doc loads <doc_root>/<slug>.md for the active release. The search index still builds from docs/, as you asked.
• Slug check I added: read_doc only accepts plain slugs (letters, digits, _, -). That stops a model-supplied slug from reading files outside the doc root.
• s3.yaml: a copy of s2 with id: s3, the title "Scout lean retrieval - pages read from versioned docs/v2", and doc_root: docs/v2. I didn't touch s1 or s2.

Eval: the 3 s3 misses come from read_doc failing for lack of docs/v2. The run used the live model, not the offline brain, because LITELLM_API_KEY is set. The three misses showed up as escalated, and the two I saw were the free-plan event cap and the annual refund. The other 15 passed from search snippets alone. Once the docs team lands the move, re-run uv run --project backend python agents/scout/scout.py eval --release s3. It should go back to 18/18.

Search index: it still comes from docs/. When the pages move, docs/*.md will be gone and the index will have nothing to search. Pointing the index at the same doc_root is a small follow-up.

I didn't run deploy s3.