allow genetic_relatedness_vector to take non-sample nodes

[petrelharp]

This should entail no changes to the algorithm. The proposal is to add an option, nodes=None, which is like:

This method can compute relatedness to non-sample nodes: if `nodes` is not None,
then `nodes` should be a list of node IDs, and the `j`th row of output will correspond 
to `nodes[j]`. More concretely, if :math:`\bar{C}_{nu}` is the genetic relatedness 
between node `n` and sample `u`, defined in just the same way as above except with 
rows indexed by all nodes `n` instead of only samples, then this method extracts those 
rows of :math:`\bar{C} W` corresponding to `nodes`.
If `nodes=None` (the default), this is equivalent to `nodes=list(ts.samples())`.