Adds a complete Cloudflare-native code-intelligence backend alongside
the existing local CodeGraph indexer.

Architecture:
- GitHub Action (codegraph-index.yml) runs the existing CodeGraph
  indexer and uploads the SQLite index to Cloudflare D1
- Cloudflare Worker (cloudflare/) serves an MCP HTTP endpoint that
  queries D1 for symbols, callers, callees, impact radius, etc.
- GitHub webhook handler auto-creates the workflow file in newly
  installed repos (GitHub App support)

Setup in three commands (from cloudflare/):
  npm run setup                   # deploy Worker + set secrets
  npm run add-repo owner/repo     # onboard a repo via gh CLI
  npm run configure-mcp owner/repo # write MCP config to ~/.claude/settings.json

Cloudflare resources:
  D1 database: codegraph (1555e587-2b51-4f85-b3e4-691f3dcaecd4, EEUR)

Co-Authored-By: Claude Sonnet 4.6 <[email protected]>

…ret put

- d1Batch now uses /query (REST API has no /batch endpoint)
- Workflow: codegraph init before index, codegraph.db not index.db
- Workflow: correct cloudflare/scripts/ URL path
- Workflow: remove npm cache (breaks non-Node repos)

Co-Authored-By: Claude Sonnet 4.6 <[email protected]>

Rewrote upload_py.py to batch multi-row INSERTs within D1's hard
100-parameter-per-query limit. Reduces API calls from 75K to ~6K
for a large repo, cutting upload time from ~50 min to ~4 min.
Adds size warnings when repos exceed 25K nodes / 50K edges.

Co-Authored-By: Claude Sonnet 4.6 <[email protected]>