Fix Kilo OAuth2 500 errors with retry logic:
- Root cause: Vercel edge node cdg1 returning 500 errors
- Added retry logic with exponential backoff (3 attempts)
- Logs Vercel edge node ID to track which node handles request
- Retries automatically route to different edge nodes
- Delays: 1s, 2s, 4s between retries

Add detailed request/response logging for Kilo OAuth2:
- Log exact request URL, headers, and body
- Log httpx version being used
- Log complete response status, headers, and body
- This will help diagnose the 500 error difference between local and remote

Add explicit headers to Kilo OAuth2 device auth request:
- Added Content-Length: 0 header
- Added User-Agent header
- Attempting to fix 500 error on remote servers

Fix Kilo OAuth2 device auth for remote servers:
- Changed from data={} to explicit content=b'' with Content-Type header
- This ensures proper form-urlencoded POST request format
- Fixes 500 error when calling from remote servers behind nginx

- Added version number to debug startup message
- Fixed Kilo OAuth2 device auth endpoint (form encoding)
- Fixed dashboard save endpoints (rotations/autoselect)
- Fixed analytics page template syntax error
- Fixed Kilo provider model prefetch at startup
- Added favicon to PyPI package

Fixes:
- Kilo OAuth2 device auth endpoint now uses form encoding instead of JSON
- Final fix for Kilo device auth 500 Internal Server Error