Add MQTT webhook bridge with resilient MQTT connection handling

FastAPI service that subscribes to nearle/# on the MQTT broker and forwards
rider status, periodic logs, and profile updates to the Jupiter webhook API.

Includes fixes for a failure mode that took the bridge down silently for
~3.8 days in June 2026, during which it served HTTP 200 on / while
forwarding zero events:

- Retry MQTT connection with backoff instead of letting a single failed
  connect kill the background thread permanently.
- Subscribe inside on_connect so subscriptions are restored after an
  automatic reconnect; the session is clean, so a reconnect previously came
  back subscribed to nothing.
- Add /healthz returning 503 when MQTT is disconnected, so a monitor can
  distinguish a live bridge from a dead one.
- Replace thread-per-message forwarding with a bounded worker pool, which
  keeps memory flat when the upstream API stalls on its 5s timeout.
- Pin requirements.txt to the versions verified in production; it was
  unpinned and had already drifted to paho-mqtt 2.1.0.
- Set PYTHONUNBUFFERED so container logs are not block-buffered.
- Name the compose image nearle-api-image to match the deployed tag.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

docker-compose.yml

@@ -0,0 +1,31 @@
+services:
+  nearle-api:
+    build: .
+    # Pinned so `docker compose up -d --build` rebuilds the SAME tag the running
+    # container was created from (it was started with `docker run`, not compose).
+    # Without this, compose would build `nearleapi-nearle-api` and leave a
+    # confusing second image behind.
+    image: nearle-api-image
+    container_name: nearle_api
+    restart: unless-stopped
+    ports:
+      - "8000:8000"
+    environment:
+      - MQTT_BROKER=66.116.225.226
+      - API_STATUS_URL=https://jupiter.nearle.app/live/api/v1/utils/createriderstatus
+      - API_LOGS_URL=https://jupiter.nearle.app/live/api/v1/utils/createriderperiodiclog
+      - API_PROFILE_URL=https://jupiter.nearle.app/live/api/v1/utils/createuserredis
+    # /healthz returns 503 whenever the MQTT side is down, so `docker ps` shows
+    # "unhealthy" instead of the container looking fine while forwarding nothing.
+    healthcheck:
+      test: ["CMD-SHELL", "python -c \"import urllib.request,sys; sys.exit(0 if urllib.request.urlopen('http://127.0.0.1:8000/healthz', timeout=3).status==200 else 1)\""]
+      interval: 60s
+      timeout: 10s
+      retries: 3
+      start_period: 30s
+    networks:
+      - nearle-net
+
+networks:
+  nearle-net:
+    external: true