Improving query -> retrieval flow

[hannahhoward]

Currently, our flow for each retrieval is as follows:

• query every candidate, wait for results back from all, with a 1 minute failure timeout.
• filter/sort candidates
• try retrievals sequentially

The downside here is we are holding for up to a minute before executing any retrievals.

A better flow could be:

• as soon as we have 1 query candidate back, start a retrieval (maybe keep other candidate queries open to get backups)
• if the retrieval succeeds, cancel any open queries and return
• if a retrieval fails, sort any queries with responses and start a retrieval on the best one
• repeat till success or all queries exhausted