Refactor health checks

[sagikazarmark]

Is your feature request related to a problem?

Since the health check actually relies on the state of underlying storage, it's not suitable for both liveness and readiness check in Kubernetes.

Describe the solution you'd like to see

Add separate liveness and readiness health checks.

Move health checks to the telemetry server, because it's only needed internally.

Liveness check should always just return "ok", readiness check should be what the health check is at the moment.

Consider using go-sundheit (beware that it imports a couple heavy/unnecessary dependencies). It's also in line with #1386 and #1397

Describe alternatives you've considered

Additional context

#853 points out that the current health check causes Dex to fail when the underlying storage is unavailable. Separating liveness and readiness checks would definitely be an improvement, but might not be enough to improve that situation. Although readiness checks should clearly signal when a pod is unable to serve traffic (and in case of a failing/unavailable storage that is the case), Dex is not HA at the moment (there are some GC jobs internally that should not run in parallel), so taking the single pod out of traffic would make Dex become completely unavailable.