fix(provider-proxy): stop one slow status probe from restarting the only replica

[baktun14]

Why

provider-proxy-mainnet runs a single replica, and both of its probes call /status with a 5s timeout and failureThreshold: 1. One answer slower than 5 seconds is enough for the kubelet to restart the container, and with one replica that takes the whole service down until the new container is ready.

That happened on 2026-09-23 at 15:21 UTC and again on 2026-09-25 at 10:26 UTC. Both times the container got SIGTERM in place: the pod was not replaced, nothing else in the cluster restarted, and the process shut down cleanly with exit code 0, which is why the restart shows up as Completed. While it was down, console-api could not reach provider-proxy at all. Every lease created in that window failed its provider reachability check with a 500 and paged the 5xx alert, and every open log, event and shell stream dropped.

The /status handler also gets slower the longer the process runs, because it walks the stats of every websocket opened since startup (2 ms on average right after a restart, 137 ms after two and a half days). That part is fixed in akash-network/console#4055, and akash-network/console#4056 runs two replicas in prod mainnet. A single slow answer still shouldn't cost the only replica.

What

• Read replicas from values ({{ .Values.replicas | default 1 }}, the same pattern console-api and provider-inventory use) and add replicas: 1 to values.yaml, so sandbox and staging keep one replica. Prod mainnet will set 2 from the console repo's .helm values.
• The liveness probe restarts the container after 3 failed checks in a row (30 seconds) instead of 1.
• The readiness probe takes a pod out of the Service after 2 failed checks instead of 1.
• Bump the chart to 0.7.7 and move the nodejs-app pin from 0.4.1 to 0.4.2, refreshing Chart.lock. With the old pin helm dependency build fails with can't get a valid version for dependency nodejs-app, so the bumped chart would not be published. 0.4.2 only adds the optional extraNodeOptions, which renders nothing when it isn't set.

Verified with helm lint and helm template. With default values the rendered manifests differ only in the two failureThreshold values, and --set replicas=2 changes only spec.replicas.