feat!: remove meta docs — the knowledge layer moves to little-memory

[AbsoluteMode]

Summary

Removes meta docs entirely. The idea — action/bug/decision cards distilled from sessions — is being rebuilt in little-memory together with the part meta docs never had: delivering a card to the agent at the right moment instead of waiting for it to search.

• Drops session_recall.metadocs (12 modules), the session-recall metadocs command, its feed into index, and tests/test_metadocs.py.
• acquire_lock was the hub indexer's only dependency on the package; it moves to session_recall/hub/lock.py (its test moves to tests/test_hub_lock.py).
• Existing indexes still hold source="metadocs" rows that nothing would ever re-index or prune. Store._drop_retired_sources deletes them (chunks, vec, fts, indexed_files) on the first writable open; afterwards it is read-only probes.
• README sections in EN/ES/ZH/RU and their .project-docs allowlist entries are removed. The user's meta docs repo is not touched — it is data, not code.
• Version 0.6.1 → 0.7.0 in all three plugin manifests and pyproject.toml: a CLI command is gone.

Why: docs/decisions/2026-10-12-remove-metadocs.md. In short: an entry warned about a missing sqlite3 on a server, and at least 6 later sessions of that project still hit sqlite3: command not found. The entries were too coarse (session-level tasks, no tool calls), and assigning a project by cwd misfiled 3 of 24 entries in one project.

Test plan

• pytest -q: 439 passed (478 before, minus the 41 meta docs tests, plus 2 new)
• New test_rows_of_removed_meta_docs_are_dropped_on_open fails without the purge and passes with it
• session-recall --help no longer lists metadocs; session-recall metadocs … is an argparse error
• After merge: git pull on the live checkout, run session-recall index, check that the 266 local metadocs rows are gone and recall still answers

🤖 Generated with Claude Code