feat: add Cloudflare Workers AI fallback for kimi models

feat: non-streaming auto-continue, higher timeout, default max_tokens

feat: add PROXY_API_KEY validation

fix: detect stream completion without [DONE] marker

feat: auto-continue + keepalive for streaming

fix: move stats to completions handler for shared /tmp access

fix: remove setKeyTotal import (read from env now)

feat: API key stats, error capture, key states tracking

feat: dashboard with stats, recent requests, console log, login

feat: proper SSE streaming via pipe for stream:true

refactor: read body in proxy, return text directly

fix: key rotation - zero delay on 402, 55 retries, handler timeout 55s

perf: reduce timeouts to fit Vercel 60s limit

feat: key rotation on retry - rotate key per attempt

fix: require paths to api/lib/

init: kimchi proxy with multi-key rotation