Connection pool management

[danielballan]

In the wild, we have seen transient errors due to a stale connection in the SQLAlchemy PostgreSQL connection pool. I think we should be using pessimistic connection pooling. The causes SQLAlchemy, when checking a connection out of the pool, to verify that it is active before using it.

I looked into this back when drafting AuthN years ago, and at least tried making it configurable. I now think we should just turn this on (pre_ping=True), unconfigurably.

[danielballan]

Related: the pool size and overflow is configurable for the AuthN database, but we have not exposed these parameters for the other databases. There is some evidence that larger pool sizes would help with performance; these should be exposed so that we can experiment.

[danielballan]

The pool sizes are configurable via #1102. Let us leave this open and consider whether we need pessimistic connection pooling. Evidence is unclear.