Fix MCP session 404s: reduce to 1 gunicorn worker

MCP sessions are in-memory per worker. With 2 workers, requests
round-robin and ~50% hit the wrong worker, returning 404. Claude Code
retries but this causes intermittent disconnects.

1 worker is sufficient for 5 users. Blue-green restarts handle
zero-downtime deploys. Also fixes misleading _connect_garmin hint.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: a-deal

gunicorn.conf.py

@@ -20,9 +20,10 @@
 # Uvicorn workers for async FastAPI
 worker_class = "uvicorn.workers.UvicornWorker"
 
-# 2 workers: enough for our traffic, allows graceful HUP reload
-# (new workers start before old ones drain)
-workers = 2
+# 1 worker: MCP sessions are in-memory per worker, so multiple workers
+# cause 404s when requests round-robin to the wrong one. Our traffic
+# (5 users) doesn't need concurrency. Blue-green handles zero-downtime.
+workers = 1
 
 # Seconds to wait for in-flight requests before force-killing old workers
 graceful_timeout = 10