Reduce h's dependence on Elasticsearch

[seanh]

We think h could use Postgres for a lot of what it currently uses Elasticsearch for, making Elasticsearch a less crucial dependency, meaning that Hypothesis developers need less Elasticsearch knowledge, and making h simpler overall.

See also: https://github.com/hypothesis/bouncer/issues/645

[mpr1255]

The tantivy plugin (basically https://github.com/paradedb) could be interesting in that regard. Brings very powerful full text search to postgres.