See http://semver.org/

Because we don't know how to name them anyway. We also ignore genes
without any transcripts.

The LRG model allows for more than one CDS per transcript (and each CDS
can be coupled with more than one transcript). The Mutalyzer gene model
does not allow that.

Previously, multiple CDSs per transcript where treated incorrectly (I
think), so now we just ignore any CDS except the first per transcript.

Unfortunately, it is a bit more involved to give good warnings for this,
so that is not implemented.

The web interface for the syntax checker previously automatically
removed comma characters from the description before parsing it. This
is not a good idea, mainly because comma characters are actually
allowed by the nomenclature (to distinguish different molecules on
RNA level).

Furthermore, this behaviour was a special case for the syntax checker
website interface and not present in either the batch or webservice
interfaces, nor in the name checker or any other interface.

This is quite a large commit, touching many things related to developer
documentation. It is all focussed on getting as much of this as possible
into the new Sphinx-based documentation.

Some highlights:

- Start Sphinx-based developer documentation, including fairly complete
  instructions for installation and configuration.
- Remove epydoc API docs.
- Rework some docstrings to conform to reStructuredText, so they can be
  used in the API docs generated by Sphinx.
- Move all of the top-level text files to reStructuredText so they can
  linked from the Sphinx-based docs and for consistency.
- Remove many obsolete things from the extras/ directory, including old
  installation scripts and migrations.

Many of the installation related documentation and scripts are removed
or adapted in light of the new automated deployment using Ansible.

Because we don't know how to name them anyway. We also ignore genes
without any transcripts.

The LRG model allows for more than one CDS per transcript (and each CDS
can be coupled with more than one transcript). The Mutalyzer gene model
does not allow that.

Previously, multiple CDSs per transcript where treated incorrectly (I
think), so now we just ignore any CDS except the first per transcript.

Unfortunately, it is a bit more involved to give good warnings for this,
so that is not implemented.

The name checker supports reverse complement ranges in insertions
and insertions-deletions, for example `3_4ins8_12inv'.

Reverse complement range insertions and insertion-deletions are not
part of the current HGVS nomenclature, but will be proposed.

The name checker supports ranges in insertions and insertion-
deletions, for example `3_4ins8_12`, and compound insertions and
insertion-deletions, for example `3_4ins[ATC;8_12]`.
The inserted sequences are accepted and concatenated before any
further processing, so reported descriptions show only the
concatenated sequences.
The support for ranges is limited to genomic descriptions.

The position converter supports compound insertions and
insertion-deletions, not ranges.

Compound insertions and insertion-deletions are not part of the
current HGVS nomenclature, but will be proposed.

For some reason, `pkg_resources.resource_filename` returns a relative path
if we call Mutalyzer using `python -m mutalyzer.entrypoints.*`. The path
is of course valid, but we pass it to Flask, which assumes relative paths
to be relative to the Python package root. To circumvent this incorrect
assumption, we first make the path absolute before passing it to Flask.