pipeline mode

Currently we send 2 queries for each request, one for the http context(+ role auth + search_path) and another one for the crud operation.

Removing the http context query grants about 33% increase in TPS with plain pgbench tests.

The 2 queries cannot be merged into one because the crud query needs the role + search_path setting beforehand.

I believe libpq pipeline mode could help us gain perf here. With pipeline mode we wouldn't need to wait for the result of the first query(which we don't care) before sending the second.