This is the butler functionality that will finally let us avoid the
death-by-many-tiny-queries performance problem in QuantumGraph
generation for (in particular) ISR.

This will allow reference catalog queries in QG generation to be
vectorized as long as they use the common skypix system.

Just had to debug some of these, and found the integer primary keys
embedded in the old messages hard to use.  Unfortunately I couldn't get
rid of all of them: this code has no access to the general mapping
from dataset type ID to dataset type name.  But I think that's the
least likely field to be causing the conflict here, so it's not a huge
loss.

Make use of the serialization caches when calling `to_simple` on
dimension records.