DEV: make sure all priors return float when needed

For a long time, some prior classes have cast inputs to numpy arrays this MR removes that behaviour and adds a test to make sure that sampling and evaluating the CDF do not cast to array.

This caused an error when evaluating the CDF many times and then combining the results.

I also streamlined some other behaviour in applying prior constraints that were originally flagged by @cjhaster.

Finally, there was an issue with sampling from joint distributions where the results could be returned out of order.