Change snapshot server to return a sqlite database

[gbrail]

In the case that the snapshot server produces a large amount of output, it won't do to return a big JSON blob. We added support to return a stream of protobufs, but this doesn't work either unless the consumer is written very carefully to process that without additional streaming.

A more scalable approach might be this:

1. Have the snapshot server produce a sqlite database containing the requested data, with a standard set of mappings from Postgres to SQLite types.
2. Have the snapshot server simply stream that result from disk.
3. Clients can use the sqlite database directly if they wish, modify it, or read it and replace it with something else.