120 model collection

[jgallowa07]

Major Changes:

• Adds initial multidms.model_collection module with multidms.fit_models for the ability to fit multiple models across a range of parameter spaces in parallel using multiprocessing. This is inspired by the polyclonal.fit_models function.
• Adds the ModelCollection class for split-apply-combine interface to the mutational dataframes for a collection of models
• Adds two altair plotting methods to ModelCollection. (1) mut_param_heatmap for visualizing aggregated parameter sets across fits, and (2) mut_param_traceplot making trace plots across fits with variable lasso coeff strengths

To do:

• Make a notebook with descriptive example of the new interface
• Convert the multidms.plot.mut_shift_plot into a method of ModelCollection for querying and visualizing aggregated groups of the collection
• Make an altair trace visualization method for ModelCollection class. this could be both trace "lines" for comparison of continuous hyper-parameters like lasso penalty coeff, as well as boxplots for sets of fits with unordered hyperparameters (different non-linearities, perhaps?)
• Cleanup and integrate notebook into documentation.
• Squash commits and final cleanup

Questions:

1. The lineplot and heatmap method throws out mutational information about "times seen". It could be useful to keep this as a slider attribute in the altair chart. However, there are several problems that arise if we want this: (1) How should times seen be aggregated across datasets (sum, mean)? (2) Should there be a slider for times seen slider for each condition (experiments), or should the times seen stat be aggregated across conditions as well?

Other Minor changes:

• removes utils module.
• Cleans up progress bar creates dependency on ipywidgets #114
• optionally removes "wts", "sites", "muts" from the mutations dataframe returned by Model.get_mutations_df. Those were unnecessary IMO
• Changes the naming of columns produced by Model.get_mutations_df(), in particular, it moves the condition name for predicted func score to be a suffix (as with shift, and time_seen) rather than a prefix. e.g. "delta_predicted_func_score" -> "predicted_func_score_delta".