Adding truncated normal distribution to `astropy.uncertainty` interface

### Description

Hello,

It would be nice to be able to use a truncated normal distribution in the uncertainty interface, for quantities that cannot go above or beyond a specific limit (for example an orbit eccentricity). I took a quick look at the interface and adding a distribution does not seem too hard, so I could try to send a PR for this if there is interest. However, NumPy does not have a truncated normal generator, so the two available options would be to use SciPy or implement it from scratch. The SciPy options seems to be the simplest, so I was wondering if it would be OK to require SciPy for this specific distribution ?

Thanks!

Start by reading the existing astropy.uncertainty interface and how its current distributions are exposed. Decide whether this distribution should use SciPy or a from-scratch implementation, then verify that a truncated normal distribution is available through the interface with the intended limits.

Written by the indexing model from the issue text.