Metrics for exemplar trajectory selection

[JosephCottam]

Choosing trajectories to show is important. We need ways to select both exemplars and outliers to show (in addition to the distributions).

Some things to look at:

• Closest to the median of the distribution
• Balance of median & variance
• Some information theoretic measure to select a trajectory that tells you the most about all other trajectories
• Existing work in time-series analysis

[JosephCottam]

So the math/code for selecting as shown in commit ef48ff31b277f4ed260c2e154d014cf8aaa1c16d is a good way to do it.

Please move it out to a utility function so that we can do something like:

trajectory_data = ... # Some dataframe
selected_traces = plots.select_traces(trajectory_data)
plots.trajectories(trajectory_data, traces=selected_traces)

There are probably things required around NAMING the traces that still need to be worked out so they don't collide with the trajectories.