Define requirements for efficient sequence feature datastore

Currently, sequence features are added to a sequence as a list of objects (ordered as added). Sequences can acquire large numbers of features (e.g. variants on genomic sequence, or heavily annotated peptide). Retrieval of features for specific criteria has to search the whole list, which does not scale well.

We may want to trade off write performance for improved read performance (by holding features in a suitably structured data store).

Before designing such a data store we need to identify the use cases it should support. That's what this issue is for.