[Feature Request] Add univariant distributions with numerical normalization

[mmikhasenko]

I have a package
https://github.com/mmikhasenko/NumericalDistributions.jl
that defines the Distributions.jl interfaces for a distribution given by an unnormalized function.

See discourse post

I'd be happy to merge it into Distributions.jl.
What are opinions, requirements? - Looking for volunteers to review PR if we agree.

How it works now in NumericalDistributions.jl:

• QuadGK for normalization, integral is stored in a struct. Btw: works for forward diff
• sampling is done by inverting CDF numerically. It's quite fast with findfirst. Pre-computation of CDF is not done in constructor, but moved to rand.
• methods mean, std, quantile are not implemented, but would be easy to add with the same QuadGK.

ping @ararslan to follow up on slack discussion.