Change distribution structure to allow vectorised calculation of methods

[mitchelloharawild]

Currently the package calls r/p/d/q functions one distribution at a time. If the distribution class handles its own indexing (rather than using a list), we can take advantage of the vectorised nature of these methods. This would improve both speed and space efficiency, but would be substantially more complicated to implement. It would also allow distribution methods to accept inputs that aren't easily split by new_dist() (such as arrays).

[mitchelloharawild]

This is lower priority now that some form of vectorisation (#52) is implemented. The value proposition here is less enticing than the complexity it introduces when compared to #52.