Replaced calls to uuid package with a queue-based labelling system in the Dirichlet Process.  Here, we create as many labels as there are data point, then pull labels out of the "pot" as we need, and replace them after use.  This allows us to remove the extra randomness in the DP process caused by the uuid generator (quite rightly) being independent of R's PRNG.

Changed ddirich to calculate on log scale rather than logging after. Changed starting tuning values for r's. Fixed lambda_j_prop calculation (didn't work if lambda j not calculated first)

Small change to PoissonNode to deal with cases when the length of the mean value is different to the offset.  Usual R conventions apply here.

Implemented AdaptiveLogDirMRW2 and DirichletNode2 which reparameterises the Dirichlet distribution as independent Gamma distributions.

Tidied up, and documented DAG and MCMC framework.  model.R contains the SA model, built using the DAG.  test.R contains test functions.

Added AdaptiveLogDirMRW for updating Dirichlet process. Still not perfect, but better than simple MRW.

Converted to R6 classes.  Still having trouble sampling from the Dirichlet distribution with the R's.

Refactored code for graph-based representation of the model.  Now need to write MCMC updaters for Stochastic nodes.