Immunedb

* Exporting clones by sample now works properly.
* Memory usage and run-time for local-alignment has been reduced.
* Documentation has been cleaned up.
* Selection pressure can now be calculated at any level.

* The API has been simplified and re-organized.
* URLs for API calls now use run length encoding to specify which samples to
analyze. This fixes issues when many samples are selected and cause the URL
to be too long.
* Grouped quality scores are now properly calculated.
* Rarefaction calculations have been removed.

* Clone mutations can now have arbitrary thresholds.
* The clone overlap query has been optimized and now properly filters
functional and non-functional clones.

* `sldb_admin` has been added to simplify creating, deleting, backing up, and
restoring SLDB instances.
* `sldb_local_align` has been added for locally aligning sequences marked as
having insertions or deletions.
* `sldb_clone_selection_pressure` has been added to calculate clonal selection
pressure. `sldb_clone_stats` now only calculates mutations and overlap, but
much more quickly.
* Duplicate sequences, regardless of ambiguous characters, are automatically
collapsed during identification.
* API call `get_stats` now takes a `percentages` parameter which will return
statistics as percentages.
* SLDB no longer uses two databases and now only requires one configuration
file.
* Identification speed has been increased..
* Samples can now be annotated with an `ig_class` specifying the isotype of the
sample (e.g. IgA, IgE).
* Sequences instances are now counted at the subject level.

* Exporting clone overlap now includes selected and all samples.
* Identification will no longer fail for samples with zero identifiable reads.
* It is no longer possible to have multiple input files for one sample.
Additionally, identification will not allow sequences to be added to existing
samples.
* Hypergeometric probabilities for V-ties are now cached, greatly improving
identification performance.
* A `--trim INT`` parameter has been added to identification allowing reads to
be trimmed prior to identification.

* TokuDB has been dropped in favor of InnoDB for the purpose of easier
installation.
* Identification tests have been re-written.
* Sequences that cannot be inserted due to a field-length restriction are added
as `NoResult`s whenever possible.
* Clonal assignment now includes partial reads by default.
* Collapsing of sequences now occurs within V, J, CDR3 length buckets for
efficiency.
* Identification now looks for `D.....C` in sequences if all other anchors
fail.
* Sample-level duplicate sequences now have the correct clone ID after.
* V-match percentage is now correct for partial sequences.