fix(github): persist GitHub indexer data between CLI invocations

[prosdev]

Fixes #36

Problem

The GitHub indexer was using in-memory storage (Map), causing indexed data to be lost between CLI command invocations. Users had to re-index before every search, breaking the dogfooding workflow.

# Before (BUG)
$ dev gh index
✔ GitHub data indexed!

$ dev gh search "MCP"
⚠ GitHub data not indexed  # Data lost!

Solution

Migrated GitHub indexer to use persistent storage:

1. Vector Storage - Store documents in LanceDB for semantic search
2. State File - Persist metadata in .dev-agent/github-state.json
3. Auto-Update - Background updates when data is stale (default: 15 min)
4. Incremental Updates - Support for since parameter (future enhancement)

Changes

Core Indexer (packages/subagents/src/github/indexer.ts)

• Added VectorStorage integration for persistent document storage
• Added state file persistence (version, lastIndexed, stats)
• Implemented isStale() check for auto-updates
• Updated search to use vector storage semantic search
• Added initialize() and close() lifecycle methods

Configuration (packages/subagents/src/github/types.ts)

• Added GitHubIndexerConfig interface
• Added GitHubIndexerState interface for persistence

Agent (packages/subagents/src/github/agent.ts)

• Updated GitHubAgentConfig to pass vector storage paths
• Removed dependency on RepositoryIndexer (was unused)
• Call initialize() during agent startup

CLI (packages/cli/src/commands/gh.ts)

• Pass vector storage config to GitHubIndexer
• Use separate storage path for GitHub data (.vectors-github)

Tests

• Added comprehensive persistence tests (indexer.test.ts)
• Tested state save/load, vector storage, auto-update, statistics
• Fixed integration tests to use new config format
• All 646 tests passing ✅

Verification

# After (FIXED)
$ dev gh index
✔ GitHub data indexed!  # 36 documents

$ dev gh search "MCP"
✔ Found 10 results  # Persistence works!

State file created:

.dev-agent/github-state.json:
{
  "version": "1.0.0",
  "repository": "lytics/dev-agent",
  "lastIndexed": "2025-11-24T13:35:09.975Z",
  "totalDocuments": 36,
  "byType": { "issue": 23, "pull_request": 13 },
  "byState": { "open": 11, "closed": 12, "merged": 13 }
}

Impact

• ✅ Data persists between CLI invocations
• ✅ Semantic search with embeddings (better results)
• ✅ Auto-updates keep data fresh
• ✅ Dogfooding workflow unblocked
• ✅ Consistent with RepositoryIndexer architecture