Improve operator stability

[adwk67]

As a user of the spark-k8s-operator I want to be able to rely on my operator to discontinue reconciliation attempts and to clean up pods from finished jobs after a defined TTL.

• define retry limit for job (init-job, driver and executor)
• configure TTL for job and driver pods for clean up?
• controller stops responding for indeterminate reasons (not does not reconcile)

[adwk67]

Current implementation:

• Job times out after 600s - do we want to make this configurable?
• Job is re-tried 6 times on failure (the k8s default for backoff_limit) - do we want to make this configurable or change this to re-starts (currently set to Never)?
• there is no value set for active_deadline_seconds: should this be changed?
• the ConfigMaps for the job, driver and executors are deleted when the parent Resource is deleted, but not before (such as on completion): change this?