feat(examples): add alchemy-cloudflare chat session example

[K-Mistele]

What

A new example, examples/alchemy-cloudflare: a fold chat host on Cloudflare, deployed with Alchemy v2.

• ChatSession.ts: the Durable Object. Each fold session is one object, named by its SessionId.
  • The first send to a new ID starts the session; after that, the object resumes it from its log when it starts up.
  • send takes whenRunning: queue (the default), steer or interrupt.
• DurableObjectEventLog.ts: fold's EventLogService stored in the object's SQLite.
• Keepalive.ts: a 30s alarm heartbeat that keeps the object from being evicted while a turn runs.
• Restarts: a turn cut off by a crash or deploy gets a "continue" nudge the next time the object starts up, at most 3 times per turn.
• Api.ts / Worker.ts: HttpRouter routes over a ChatSessions service, which calls the Durable Objects.
  • POST /sessions/:sessionId/messages
  • GET /sessions/:sessionId
• Model: OpenAI gpt-5.6-terra on medium reasoning, through openaiModel. The deploy reads OPENAI_API_KEY from your environment and stores it on the Worker as a secret.

Other changes

• Root package.json: adds the examples/* workspace, and typecheck / test now also run over the examples.
• Catalog: adds alchemy, @cloudflare/workers-types and @effect/platform-bun.
• fold-core is unchanged.

Why not Workers AI

Alchemy's Workers AI LanguageModel repeats streamed text on some models and never emits streamed tool calls. fold always uses streamText, so both bugs hit it. Reported upstream: alchemy-run/alchemy#1907.

Testing

• bun run typecheck, lint and the route tests (test/Api.test.ts, against an in-memory ChatSessions) pass.
• Deployed and tested live on Workers AI: sessions start, resume and persist across requests.
• The switch to OpenAI hasn't been deployed yet.

🤖 Generated with Claude Code