Add a UserDistribution type to the list of distributions about bpswg HOT 4 CLOSED

Elsewhere we used the term "non-parametric." I understand "non-parametric" to 
be a distribution provided by a collection of points. I think we have the same 
thing in mind, but are representing it differently. I think your weights are 
based on the frequency of occurrence of that point ??? We should settle on a 
terminology.

Another issue, though I think counting ("bins" in histograms) is fine in 
discrete valued spaces (integers) I'm not sure whether counting real valued 
data is a good idea unless you give ranges.

Done in version 0.2

The UserDistribution class provides a custom sampling of points with the 
likeliness of each one to occurs.

The UserDistributionDataPoint class represents a data point in the User 
Distribution

At the session on the 23rd Feb there was discussion about the parameters for a 
user distribution. The initial suggestion was that the weights could be 
expressed as real data without normalisation (i.e. dont have to sum to 1 or 
100). In our experience this is user friendly as it allows empirical data to be 
entered. Tools would then handle these weights. 

Whilst we can demand that the weights sum to 1.0 in reality this isnt 
absolutely necessary.