Notes on improving boilerplate for distribution objects:

• creating boilerplate layer - this should encompass broadcasting, index/column subsetting at the start of methods.
• we should think about dynamic docstrings to contain explicit formulae for different distribution methods, where available. Possibly a dict with math formulae, and the same approach as set_config in sktime.
• it would be good to also support parameter/index cases for scalar and multivariate single-instance. In the first case, the behaviour should return scalars, in the second, 2D tables.

Also for discussion.