Analysis filters apply across all transcripts (rather than only representative)

At the moment, we only filter against the "representative transcript" (ie VEP pick) with a special case for handling overlapping genes (for gene lists etc)

It would be good for filters to work against all transcripts. Trouble is, there are 49,594,389 transcript annotations in VG test and 8,811,654 variant annotations => ie 5.6x transcripts/variant on average.

A possible solution is to make a special "most damaging entry for each field" which we filter on. The issue is then, if we show that ,the transcript isn't real. If we show the normal representative one, people will wonder why it returned any results.

Start by reviewing how filters currently use the representative transcript and how overlapping genes are handled. Use the VG test annotation counts as a scale reference, then define how filtering across all transcripts should choose a damaging entry and how results should present the displayed transcript.

Written by the indexing model from the issue text.