Split Reference.accession into accession and version

[martijnvermaat]

It's a bit annoying that the optional version number is included in the Reference.accession column. For one, it makes querying by just accession (without version) hard.

Plan of action for migration:

• [x] Add Reference.version column. #400
• [x] Update codebase to write versions to Reference.version (in addition to Reference.accession). #400
• [ ] Release.
• [x] Populate Reference.version using Reference.accession. #401
• [x] Add unique index on Reference.accession and Reference.version. #401
• [x] Drop unique index on Reference.accession. #401
• [ ] Update codebase to read from Reference.version and ignore potential versions in Reference.accession. #401
• [ ] Update codebase to not write versions to Reference.accession.
• [ ] Release.
• [ ] Remove versions from Reference.accession.
• [ ] Update codebase to remove ignoring potential versions in Reference.accession.

[martijnvermaat]

While working on this, I came to the conclusion it's not a good idea to split accession and version. It introduces a lot of complexity for little benefit.

In general, Mutalyzer always sees 'accession.version' as the identifier of the reference and because we always want exact identifiers, there is little need for accession numbers without version.

The most obvious use case I see for a split is that we can easily query available references with a certain accession, not taking version into account, as a way to inform the user when a specific reference identifier was not found. But I guess we better have this use case as the exception, and make our life easier for the rest.

So I guess I'm aborting this for now. Addition of the version column has already landed in the master branch, but this is easy to roll back. The original column has not yet been touched in master.

[martijnvermaat]

Reverted in #402