Null values

[skhilton]

This pull requests handles null values in the data. There are three main changes to the data to handle null values

1. data cleaning
   During the data cleaning step, metric values which are undefined, null, NaN, false, or "" are all cast to undefined.

2. circles disappear
   If the metric value is undefined, the circle radius is set to zero.

3. The protein structure turns grey.
   If the metric value is undefined, the point remains highlighted on the protein structure but turns grey to signal missing data. We could also consider removing it from the protein structure all together.

The idea behind these changes is that we don't actually remove the point so it is still classed as selected. When you change to another metric, the point reappears with the correct color.

I don't believe that any changes need to be made to handle null values for the logoplots but if you could check that @huddlej that would be great!

An improvement we could make in the future is to have a metric option disappear from the dropdown menu if none of the sites have valid values for the metric. This is described in issue #70.

Closes issue #69

[skhilton]

whoops, @huddlej, forgot to mention. I have included a test data set in _data/test.csv. There are no valid values for VIDD5 mindiffsel. This file should probably be deleted before the changes are merged.

[huddlej]

This works for me! There was a minor issue where the context view tried to draw a path with "NaN" values and I've added a commit to address that.

The test data were very helpful. I think we should keep these somewhere in the repo, if not in the top-level docs directory, for future testing.