Priority Task Processor Improvements

[yycptt]

What changed?

1. Fix nil pointer exception when changing dynamic config for enabling priority task processor. This only happens when a shard is reloaded after the config is changed but before the host got restart.
2. Update task processing retry policy to have a max attempts of 3 before putting into redispatch queue. This is for preventing standby tasks from occupying a worker goroutine for too long. For active tasks, in most cases the task will be completed before its third attempt.
3. Update the logic for task priority assignment. Idea is to assign high priority to no-op tasks (tasks for standby domain submitted by active processor or vice versa) so that they won't get a low priority and block the ack level (especially for active processor) from advancing.

Why?
Bug fix and performance improvements.

How did you test it?
Unit test. Local canary.

Potential risks
N/A, feature flag disabled

[coveralls]

Coverage increased (+0.05%) to 68.398% when pulling 4948fdc on yycptt:improve-priority-processor into da4cffa on uber:master.