BigFloat support

[wsmoses]

see https://github.com/LeilaGhaffari/my-enzyme/issues/2

cc @LeilaGhaffari @jedbrown

[wsmoses]

@LeilaGhaffari per your other Github issue, one what timeline would support for this be needed?

To do some expectation setting / figure out prioritization, enabling support for custom base types (like this) are on the roadmap, but do have a nontrivial amount of infrastructure that would be required first so it's not an instantly solvable thing. For BigFloat in particular, we may be able to just use custom derivative rules for the various operators and that may be sufficient -- but that would also be ideal to have https://github.com/EnzymeAD/Enzyme.jl/pull/177 land first so it would be easier to write all of those.

[jedbrown]

Honestly, I don't think it's a big deal. The BigFloat here is for reference values to measure accuracy of the f32/f64 computation. But we can compute reference values using Zygote or with "by hand" differentiation.

[jedbrown]

Better to put resources into #177.

[LeilaGhaffari]

Thanks for creating this issue. I went ahead and closed #676.