Add fallbacks for log1pmx and logmxp1

[simsurace]

This looks like it would make sense, given that the other functions dispatch on Real argument, and it also fixes #44.

[devmotion]

As I mentioned in #44, I think this PR is not the correct fix for #44, and hence the motivation is a bit different.

Regardless of that, I think it is reasonable to extend these functions to ::Real. My only concern is that the proposed implementation might not be sufficiently accurate for Float32 and Float16 for which we provide optimized implementations for other functions. It would be good to check that it is sufficiently accurate for these types, and in particular the numerically problematic inputs.

[simsurace]

I see. I need to look a bit into the floating-point issues here.
I might need help from someone more familiar with those things here.
In particular, I don't really understand the optimized implementations for Float64.
I expect the problematic inputs to be at different locations for other floats.

[simsurace]

Given that the PR JuliaDiff/DiffRules.jl#82 is missing this fallback, I would propose to merge this and release a new version, open a separate issue for optimizations, and add those in a future PR.

EDIT: I opened #46

[devmotion]

We don't need the Float32 support in DiffRules, it's just a part of the tests that we can skip (or check that it errors).

[devmotion]

Seems fine in principle, I guess we can optimize the generic implementation in follow-up PRs if needed.

Can you add some tests with non-Float64 types (eg some Float32 values and compare their results with Float64 results) and update the version number?

[devmotion]

LGTM, thanks!