Model initialisation

[ben18785]

Currently, the MCMC is initialised using arbitrary FOI values etc. which vary according to the specific model being fitted. To make this more robust, I'd suggest the following approach:

1. We run stan::optimisation a number of times (default is say 10), each time starting from random initial values
2. We select the single parameter set which is the best fitting amongst the above runs
3. We perform MCMC using the parameter set found in 2 as initial points

[ben18785]

We probably will also want to give the users the opportunity to perform an optimisation as opposed to sampling. But this is best handled in a separate issue.

[rccreswell]

@ntorresd I'm happy to look at this.