Improve automatic construction of transforms from prior support

[janfb]

Currently, we check for prior with interval bounds (i.e., uniform and box-uniform) prior and then build a transform to unbounded space for them.

For priors that do not have a support of type constraints.interval we alway use the mean and std (from properties or estimated from samples) to build an affine shift and scale transform.

This ignores half-bounded prior, e.g., LogNormal or Gamma. Thus, it would be great to improve and automate the checks for prior bounds to include all common types of priors.

Relates to #585

Partly fixed in #602

[janfb]

Update: this seems to be done by biject_to automatically, so no detailed checks needed.

See #619