File attachments are a bit expensive to query fully, because they may be

bound to a ReviewRequest or a ReviewRequestDraft, wither as an

"active" or "inactive" file attachment. Depending on the API and the

request, we may need to get a subset of those or all of those, and that

can lead to a lot of queries. On top of that, we needed to then fetch

the review request or draft associated with those, and this triggered up

to two more requests per file attachment.
This change improves upon all of this, minimizing queries and accesses.
We now use the relation counters on the ReviewRequest and

ReviewRequestDraft to determine which fields we want to even consider

querying. This will, in many cases, reduce the query counts.
We then build a list of fields for prefetch_related(), which will let

us query those review requests or drafts all at once, rather than

per-result.
We also avoid any exclusion filtering for the main review request file

attachment resource if we know the exclusion list will be empty.
And finally, if all considered counts are 0, we just return an empty

result immediately, rather than hitting the database.
This is still a lot more queries than we'd ideally have, but it does

keep the query counts at an upper limit, rather than being proportional

to the number of file attachments returned.

Unit tests pass.