Dms-variants

------

Added
+++++++
- The new ``AbstractEpistasis.single_mut_effects`` method.

- Options ``returnformat`` and ``stringency_param`` to ``AbstractEpistasis.preferences`` and ``utils.scores_to_prefs``.

Changed
+++++++
- ``AbstractEpistasis.preferences`` and ``utils.scores_to_prefs`` return site as integer.

------

Fixed
++++++
- Errors related to using ``pandas.query`` for ``nan`` values. Not sure of the cause, but the errors are fixed now.

------

Changed
++++++++
- Eliminated the default log base for conversion of scores / phenotypes. This is because base 2 gave excessively flat preferences, and the choice of a base is something that the user should need to think about. Added explanation about the consequences of this choice to docs and examples.

- The preferenes returned by ``scores_to_prefs`` and ``AbstractEpistasis.preferences`` are now naturally sorted by site.

------

Added
++++++
- The new ``AbstractEpistasis.preferences`` method gets amino-acid preferences from phenotypes.

- Added ``utils.scores_to_prefs``.

------

Fixed
++++++
- The ``isplines`` module now uses a simple dict-implemented cache rather than ``methodtools.lru_cache``. This fixes excess memory usage and allows objects to be pickled.

- ``AbstractEpistasis`` internally clears the cache via ``__getstate__`` to reduce size of pickled objects. This avoids pickled models being huge. Also added the ``clearcache`` option to ``AbstractEpistasis.fit`` to serve a similar purpose of memory savings.

--------

Added
++++++
- Added additional forms of likelihood function to the global epistasis models. This involves substantial re-factoring the epistasis models in ``globalepistasis``.
In particular, the ``MonotonicSplineEpistasis`` and ``NoEpistasis`` classes no longer are fully concrete subclasses of ``AbstractEpistasis``.
Instead, there are also likelihood calculation subclasses (``GaussianLikelihood`` and ``CauchyLikelihood``), and the concrete subclasses inherit from both an epistasis function and likelihood calculation subclass.
So for instance, what was previously ``MonotonicSplineEpistasis`` (with Gaussian likelihood assumed) is now ``MonotonicSplineEpistasisGaussianLikelihood``.
**Note that this an API-breaking change.**

- Added the ``narrow_bottleneck.ipynb`` notebook to demonstrate use of the Cauchy likelihood for analysis of experiments with a lot of noise.

- Added the ``predict_variants.ipynb`` to demonstrate prediction of variant phenotypes using global epistasis models.

- Added ``simulate.codon_muts``.

Fixed
++++++++
- Some minor fixes to ``codonvariat_sim_data.ipynb``.