Need simple improved queueing for arriving queries

Over and over we have issues related to arriving queries and blockages (stuck-ness behind long running queries). We really need to introduce a set of input queues (rather than one queue) that can be properly used to manage this (e.g., metadata and status queries would be separately queued) as well as to have parity with competitors like Redshift that support multiple queues and simple query request tagging to do multi-class load management. (Shiva's PhD work will be relevant here.)