Parallelization options

[jgabry]

Currently when the user selects the number of chains we use that number as the number of cores to use both for MCMC and for LOO. Some potential future improvements could be to:

• use within chain parallelization (whether this is feasible and beneficial will depend on the model)
• use as many cores as possible for LOO (in theory LOO can use as many cores as there are observations in the data)

This is not the highest priority, but something we could think about in the future.