feat(core): Incremental indexing via rebuild strategy

[prosdev]

Summary

Fix the existing update() method to actually work. Most infrastructure already exists.

What Already Works

• ✅ State management with file hashes (indexer-state.json)
• ✅ detectChangedFiles() using SHA-256 comparison
• ✅ update() method orchestration
• ✅ Doc IDs contain file path (e.g., src/auth.ts:func:10)
• ✅ CLI dev update command

What's Broken

1. delete() throws error — We just never implemented it
2. New files not detected — Only iterates state.files, misses new files
3. Deleted files leave orphans — Not cleaned up

Implementation

1. Implement delete by file path prefix (~1 hour)

// In store.ts
async deleteByFilePrefix(filePrefix: string): Promise<void> {
  await this.table.delete(`id LIKE '${filePrefix}:%'`);
}

2. Add new file detection (~30 min)

// In detectChangedFiles()
// Also scan for files NOT in state.files
const allFiles = await scanForFiles(repoPath);
const newFiles = allFiles.filter(f => !state.files[f]);

3. Handle deleted files (~30 min)

// Files in state but not on disk
const deletedFiles = Object.keys(state.files)
  .filter(f => !existsSync(f));
// Delete their docs from vector store

Success Criteria

• dev update detects changed files
• dev update detects new files
• dev update removes docs for deleted files
• Orphaned symbols cleaned up when removed from file
• --force still triggers full reindex

Out of Scope

• Git commit SHA tracking (optimization, not required)
• Watch mode

Parent Epic

Part of #104 - Performance & Reliability Critical Path