Improve event loop timing metrics

SAMZA-401 added the event-loop-utilization metric to measure the % active time in the event loop. However, the value for the event-loop-utilization (at least since samza 10) is almost always ~= 1.0, even for over-provisioned jobs, which is not very useful.

The way it is currently measured, the active time includes choose-ns which includes time spent in poll() waiting for new messages. This waiting should be considered idle time.

The goals of this ticket are to:
1. Exclude the choose time from the active time to fix the event-loop-utilization metric.
2. Add deserialization-ns and poll-ns metrics to help users understand where the time is being spent in the chooser.