Refactor overlap with rand_distr to extend/mirror their implementation

[YeungOnion]

There's quite a bit of overlap with rand_distr for distributions we both implement:

• struct for distribution families
  • fallible constructors with parameter errors (many are enums of unit types, address size concerns in Consider improving error handling and StatsError type #221)
  • more consistent with their numeric generics
• impl rand::Distribution, we'll immediately get updates

Perhaps we can lean into this a little bit to establish some more consistency. Do we need a crate level error, and would relying on thiserror for transparent errors work?