Bucketing: Look for commonalities amongst timeouts

[Chris Warburton]

One obvious thing to try is to plot the frequency of certain names
amongst the total population of samples, and amongst the failing
samples.

We can then compare these distributions by eye, and possibly with some
statistical test, to see if some names are more common. If some *do*
turn out to be more common, we might look at their types to see if
there's a pattern there (for example, I imagine that functions like
't -> t' or 't -> t -> t' might cause a blow-up in the number of
combinations)