Remove registration timeout and retry logic

[erwa]

Currently, there are a couple timeouts involved in worker/parameter server registration:

• tony.task.registration-timeout-sec (default 300 sec)
• 120 sec polling until non-null in TaskExecutor.registerAndGetClusterSpec()

    return Utils.pollTillNonNull(() ->
        proxy.registerWorkerSpec(jobName + ":" + taskIndex,
            InetAddress.getLocalHost().getHostName() + ":" + rpcPort), 3, 120);

If there are large container scheduling/start-up delays, jobs can fail due to this. We should remove these timeouts entirely. We also then don't need the tony.task.registration-retry-count property either.

[erwa]

Internal JIRA: LIHADOOP-42079