Fix stale read on TranscriptionEntity corpus check

Closes #1384 (closed)

This does not add any new .using('default'), because this resolves the stale read by removing the offending queries. The query to retrieve the transcription's element now uses a select_related on a query that already uses .using('default'), and the two queries to retrieve the transcription's element's corpus and the entity's corpus have been removed because we can just compare by corpus_id.