Parallelise the cluster (regular mixture model) algorithms more efficiently

[dsteinberg]

At the moment only the 'grouped' mixtures and more hierarchical algorithms are parallelised in an efficient manner.

[dsteinberg]

This code is really slow compared to just dividing the data into groups and running e.g. the parallel G-LDA algorithm...