OpenAI-compatible Agent Y2 chat stream

Routes OpenAI-style chat completions requests to Agent Y2 and streams text responses to compatible clients. v1 supports `messages`, `stream: true`, optional `model` set to `y2-agent` or `agent-y2`, optional `threadId`, and optional bounded `metadata`. Arbitrary models are unsupported. System and developer messages cannot override Agent Y2's fixed instructions. v1 requires an API key and does not support x402. Both Agent Y2 routes share a throttle of 5 requests/minute and 100/day per key, plus 10 requests/minute and 250/day per user or workspace. A `402` indicates subscription, chat-credit, or upstream-provider credit exhaustion, not an x402 challenge.