Add Grok CLI backend and per-engine backend documentation

[tony]

Summary

• Add Grok CLI as agentgrep's fifth backend — prompt history, chat sessions, and session search index are all searchable out of the box
• Add docs/backends/ section with a support matrix and per-engine pages documenting store layouts, adapter IDs, and record schemas for all five agents
• Add 10 store catalog entries for Grok (3 with live discovery, 7 documentary-only) covering sessions, prompt history, search index, events, summaries, memory, logs, worktrees, and config
• Add integration tests for all three Grok parsers: JSONL prompt history, JSONL chat transcripts, and SQLite FTS5 session index

Changes by area

Runtime (src/agentgrep/)

• __init__.py: Register "grok" in AgentName and AGENT_CHOICES; add discover_grok_sources() honouring GROK_HOME env override; add parse_grok_prompt_history, parse_grok_chat_history, parse_grok_session_search_db parsers; wire adapter dispatch in iter_source_records()
• stores.py: Add "grok" to AgentName literal
• store_catalog.py: Add _GROK_STORES (10 descriptors), bump catalog_version to 4
• mcp/_library.py: Add "grok" to AgentName, AgentSelector, and 3 adapter IDs to KNOWN_ADAPTERS
• mcp/models.py: Add "grok" to all 5 agent Literal fields
• query/registry.py: Add "grok" to the agent field's enum_values

Tests

• tests/samples/grok/: Redacted JSONL fixtures for prompt history and chat sessions
• tests/test_agentgrep.py: 5 new tests — env-override discovery, prompt history parsing, chat session extraction, empty-content filtering, SQLite FTS5 index parsing
• tests/test_stores.py: Add "grok" to KNOWN_AGENTS so existing catalog invariant tests cover the new entries

Documentation

• docs/backends/: New section with index.md (support matrix) and per-engine pages (codex.md, claude.md, cursor.md, gemini.md, grok.md)
• docs/dev/storage-catalog.md: Add Grok CLI section documenting path layout and adapter notes
• docs/index.md: Add Backends grid card and toctree entry
• docs/dev/index.md: Update storage catalogue card to mention Grok

Design decisions

• Catalog-driven discovery: discover_grok_sources delegates to discover_from_catalog() — no custom file-walking logic. The generic machinery handles globs, file checks, and deduplication identically to all other agents.
• All record types emitted: The chat history parser emits system, user, assistant, tool_use, and tool_result records rather than filtering to user+assistant. This maximises searchable content and matches the Codex session parser's approach.
• Unix → ISO-8601 timestamp conversion: The SQLite parser converts updated_at (unix seconds) to ISO-8601 for consistency with all other adapters' timestamp fields.

Test plan

• test_discover_grok_sources_honours_grok_home_env — GROK_HOME env override takes priority over ~/.grok
• test_search_grok_prompt_history — JSONL audit log yields kind="history", role="user" with correct session_id
• test_search_grok_chat_history_session — JSONL transcripts yield user + assistant records with correct conversation_id
• test_search_grok_chat_history_drops_empty_content — records with empty/whitespace content are filtered
• test_search_grok_session_search_db — SQLite FTS5 index yields titled records with ISO-8601 timestamps
• Existing catalog invariant tests pass with grok entries (test_stores.py)
• Full suite passes
• just build-docs builds cleanly with new backend pages

[tony]

Code review

Found 2 issues:

1. pyproject.toml description and keywords omit Grok. The PR adds Grok as a fifth backend but the package metadata still lists only four agents: (Codex, Claude Code, Cursor, Gemini) in the description and ["ai", "codex", "claude", "cursor", "gemini", ...] in keywords. Users discovering the package via PyPI search won't know Grok is supported.

agentgrep/pyproject.toml

Lines 3 to 5 in bb4115c

	version = "0.1.0a7"
	description = "Read-only search for local AI agent prompts and history (Codex, Claude Code, Cursor, Gemini)"
	requires-python = ">=3.14,<4.0"

agentgrep/pyproject.toml

Lines 23 to 25 in bb4115c

	keywords = ["ai", "codex", "claude", "cursor", "gemini", "mcp", "search", "agent-history"]
	readme = "README.md"

2. README.md description omits Grok. The opening line lists "Codex, Claude Code, Cursor, and Gemini" without Grok, despite the PR adding full backend support.

agentgrep/README.md

Lines 7 to 8 in bb4115c

	Read-only search for local AI agent prompts and history across Codex,
	Claude Code, Cursor, and Gemini.

🤖 Generated with Claude Code

_{- If this code review was useful, please react with 👍. Otherwise, react with 👎.}