switch between two query planner instances

[Geal]

right now we only use one query planner instance, in one deno VM, to prevent some reliability issues. Unfortunately, that introduced another reliability issue: when the query planner loads a new schema, new queries have to wait before they can be planned.
Proposal: set up two query planner instances, one active serving queries, the other one waiting. When a new schema comes in, the waiting instance starts loading the schema while the active one serves current queries. When the waiting instance is fully loaded, they switch places, it becomes the active one.

[gwardwell]

Even if it ends up with multiple instances for real-time query planning (which I think is a must-have) I would have a dedicated instance for rebuilding the query plan cache during the schema refresh process. I think getting that decoupled from real-time operations would do a lot for preventing the latency spikes we see during schema refreshes.