Switch k8s pods to use real healthcheck liveness probe instead of restarting pods daily

[lotas]

Current implementation makes pods restart daily by telling kubernetes it should be killed.
This results in connections being dropped and increased 502 rate.

Note: 502 are not directly related to this problem and is rather a separate issue that needs to be solved around pod termination.

TODO:

• Replace livenessProbe unconditional termination with periodic healtcheck endpoint call /ping to let k8s know if pod is still responsive.
• Set memory limits to kill pods automatically in case of memory leaks (size tbd)