Numerical stability problems for autodiff with censored_pmf

[SamuelBrand1]

The draft PR #453 which recreates some analysis features of primarycensoreddist, but has run into problems due to numerical stability problems with taking AD through censored_pmf function.

A series of methods to try and address this:

• Switch numerical quadrature method to Integrals.jl so we can try more methods. #459
• Try refactoring censored_pmf to promote grad stability. #460
• Change the integration problem into an IVP problem and solve using an ODE solver; this might be better in terms of getting gradients since adjoint methods are important in ODE problems.

[SamuelBrand1]

It looks like #460 fixed this in #453.

[SamuelBrand1]

Closed by #453