CN-model calibration #337

(A)I started implementing something based on latin hypercube sampling (LHS) and committed. vignettes/example_cmodel.Rmd contains a demo but the posterior weights are too unevenly distributed across parameter combination samples. Potentially part of the solution but also other open points are:

• Drawing more than 1000 samples (eventually need to run this on the cluster with large CPU nodes)
• Some iteration to the search, densening sampling near optimum
• Solution to the large autocorrelation of errors in fluxes
• Come up with a step-wise calibration (P-model only, then C-model, then CN-model) taking posterior of one step as prior of the subsequent step.
• Multiple targets calibration also for LHS

This is now in cnmodel branch.