Add a warning if non-uniform events or non IID primary events are used in the latent model

[seabbs]

#474 made it possible to specify any priors on the primary events (and the secondary events).

In the secondary event cases there is no case where setting a different prior is a good idea and so we should error I think.

In the primary event case there are several cases but the current implementation of our posterior prediction and log_likelihood functionality doesn't support this so the output from those functions could be misleading. I think this should be a warn as the inference functionality is still useful.