[BUG] Fix the rescaling of subsets of ConditionalPriorDict and streamline subset resolving

[JasperMartins]

This PR addresses Item 3 in #1026

To reiterate, current issues for rescaling/sampling/evaluating subsets of keys are:

• Currently, if a subset of the prior-keys is requested (eg. 3 keys for a PriorDict with 4 priors) and a theta with the appropriate length (3) is passed, rescaling fails because the function attempts to rescale all non-fixed priors. Also, it would be necessary to check if the subset of requested keys can be resolved.
• The current implementation would treat any ConditionalBetaDistribution as a "fixed" prior and add it to any prior evaluation, which could lead to wrong results.

This PR addresses these issues and streamlines subset-resolving.