[bugfix] Frontend: give GQ_BACKEND_HOST a working default to fix 503s

[els0r]

A remote single-host deployment that left GQ_BACKEND_HOST unset hit
api request failed: status=503 with Caddy logging "no upstreams available":
the image's default for GQ_BACKEND_HOST was the empty string, so
reverse_proxy /_query* {$GQ_BACKEND_HOST} expanded to a proxy with no upstream
to dial.

Give the proxy a sane fallback so it works out of the box:

• Caddyfile: default the env placeholder to http://127.0.0.1:8146, so an
  unset or empty GQ_BACKEND_HOST still yields a valid upstream.
• Dockerfile: set the same value as the image-level ENV default instead of
  an empty string.
• docker-compose.yaml: the frontend default pointed at :8145 (the goProbe
  sensor API port) while global-query listens on :8146 (SERVER_ADDR) — fix
  the mismatch so the documented stack proxies to the right service.
• README: document GQ_BACKEND_HOST vs GQ_API_BASE_URL and the
  "no upstreams available" failure mode.

Closes #480

🤖 Generated with Claude Code

[fako1024]

Some comment / documentation improvements would be appreciated, aside from that requires testing, unclear if this solves the issue.

[fako1024]

A small chart displaying the interaction between the components, in particular frontend + caddy, global query and goprobe nodes (+ e.g. production environment with an nginx / TLS reverse proxy in front) would help.

[fako1024]

The comment here is still weak in terms of clarity (it should be clear that this the global query endpoint, the phrase is still too weak, even more so if an actual reverse proxy like nginx is involved in a production environment).