Rewrite epidist_gen_log_lik to more efficiently wrap brms log_lik

[seabbs]

#474 introduced a generic epidist_gen_log_lik which wraps brms functionality and so avoids the needs for manual specification of reparameterisations. Unfortunately, the approach take to do this duplicates calls to the underlying likelihood by the number of samples (i.e if there are 2000 samples each call duplicates cdf calls 2000 times). It is unclear how much of a performance hit this is or if it can be avoided.

[seabbs]

I have extensively inspected the prep object and attempted a few approaches. I think without moving to the kind of manual approach taken for #477 this may be very hard. Unless I have a revolution I will implemented the key dists manually (that don't have analytical solutions but the same approach can be used, add docs about more support and close based on that. I think potentially we may want to create a new issue for automating if we do go that path.