feat(langbot): add LangBot + Valkey vector search cookbook

[atao2004]

Summary

Migrates the archived LangBot + Valkey cookbook (from archive/main-pre-reorg) to the current CONTRIBUTING.md format, scoped to the Valkey Search vector-database backend only.

• 01 - Getting Started — start Valkey with the search module, verify the connection with valkey-glide, enable LangBot's opt-in valkey_search backend
• 02 - Vector Search (was 03) — HNSW index, embeddings as HASH keys, KNN queries via the typed ft API
• 03 - Hybrid Search & Filtering (was 04) — full-text, hybrid filter-then-KNN, file_id TAG filtering, vector_weight caveat, query-injection safety

Scope change from the original 4-cookbook series

The original series also covered distributed rate limiting (valkey_fixwin). That cookbook is dropped in this PR: the upstream implementation, LangBot PR #2267, was not merged — the maintainer said it would be deprecated in favor of a plugin-based approach ("we're going to deprecate Pipeline in the upcoming versions, so this feature will not be accepted, and we recommend implementing this as a plugin"). The vector-search backend (LangBot PR #2276) is merged and released (LangBot v4.10.6), so this PR ships that half of the series now. Rate limiting may return as a separate cookbook if/when a plugin-based reference implementation exists.

Version pins

• valkey/valkey-bundle:9.1.0 — the TEXT field used by this backend needs valkey-search ≥ v1.2.0, which ships in the 9.x bundle line
• valkey-glide==2.5.0 — latest published release; verified compatible by actually running the sample and test suite against a live container

A bug found and fixed during review

Independent review caught a real issue in the code carried over from the archive: file_id_filter() percent-encodes TAG-unsafe characters ({, }, *, %) when building a query, but the write path stored file_id raw and unencoded. A stored value containing one of those characters would silently stop matching its own filter forever — directly contradicting the cookbook's own claim that this "fails closed." Fixed by encoding at write time too, corrected the affected doc text (injection-safety section and troubleshooting table in cookbook 03), and added a regression test (test_delete_by_file_id_with_tag_unsafe_characters) that fails against the old code and passes against the fix.

Validation

• valkey-glide==2.5.0 installed and run against a live valkey/valkey-bundle:9.1.0 container
• sample/vector_search.py demo runs end-to-end; its README's "Expected output" block is captured from an actual run, not guessed
• sample/test_vector_search.py: 8/8 tests pass (7 original + 1 new regression test)
• markdownlint-cli2 clean on all line-length (MD013) findings

Pre-Submit Checklist

• Builds and runs from a clean clone
• Runs against current stable Valkey Bundle (9.1.0)
• No private links, internal references, or local paths
• No marketing language or uncited claims
• Vendor-neutral by default (LangBot itself is the framework being integrated; no third-party vendor lock-in introduced)
• README explains the Valkey concept demonstrated, prerequisites, and how to run
• Dependencies pinned to specific versions
• CI validates the sample (temporary cookbook-python.yml copy, per the pattern used by cookbook-semantic-caching/cookbook-haystack/cookbook-agno — PR docs+ci: add semantic-caching cookbook with Python CI (MERGE FIRST FOR PYTHON CI) #45's shared workflow hasn't merged yet)

🤖 Generated with Claude Code

[jbrinkman]

Closing this draft PR as part of a cookbook program cleanup.