Add sensitivity for in-sample logscore (log predictive density / LPD)

[n-kall]

In many cases, there are too many parameters in a model to consider the sensitivity of all of them. Instead a global quantity such as in-sample log score for the observations could be used, which would apply to all models.

As the log_lik is already used in priorsense, this should be quite straightforward to implement.