Search indexing was painfully slow, since every single ReviewRequest

being indexed ended up adding one query for the submitter and one to get

a list of all FileDiffs.
The FileDiffs query in particular resulted in quite a lot of unnecessary

time spent.
By taking advantage of select_related(), prefetch_related(), and a

prepare_file() statement, we can reduce this down to only 2 additional

queries per batch of review requests (as defined by Haystack, defaulting

to 1 large batch).

Ran an index against a local MySQL with 2,572 review requests, and tested that

file queries still worked as expected.
Added some debugging to the update_index command to see the resulting query

counts.
Before this change, there were 7,724 queries for 2,572 review requests.
After this change, there were 14.