Fix blocking resource groups when query queue full

[rmarduga]

When Presto Coordinator cannot run query nor queue query, it fails the error. And Coordinator does it under root resource group lock. Failing query triggers query state change that in particular triggers event listener handler that could be expensive. Doing it under resource group lock essentially blocks the resource group and severely degrades performance.

Test plan - (Please fill in how you tested your changes)
I ran the coordinator without and with the change. Within 1 hour I sent around 60K queries to the cluster. Less than 2K queries successfully executed and all other failed. Around 75% of the queries failures were due to QUERY_QUEUE_FULL error. I recorded the sessions with Java Flight Recorder. Picture below shows selected threads that were locked on RootInternalResourceGroup lock.

Here, the yellow blocks depict the thread waiting on the lock.
The picture below demonstrate the same settings with applied this code change.

Overall, the locking time for RootInternalResourceGroup monitor type reduced from 3 hour 24 min to 1 min 11 sec and for InternalResourceGroupManager monitor type it reduced from 4 days 4 hour to 9 sec.

Fill in the release notes towards the bottom of the PR description.
See Release Notes Guidelines for details.

== RELEASE NOTES ==

Resource Groups Changes
* Fix blocking resource groups when query queue full