Improve get_conversions_for_file query performance on large sites

There's a seq scan that's occurring in the query produced by `get_conversions_for_file` (files/classes/conversion.php)

For some reason, Postgres is choosing to do that rather than use one of the indexes. This query is regularly sitting in the top 4 for CPU on both the various reader and writer instances we're using.

Also I've tested this on a smaller site and the issue wasn't apparent. I've only observed this on larger sites.