add `ForwardRuleConfigBackend` (to match `ReverseRuleConfigBackend`)

[gdalle]

If I understand correctly, AbstractDifferentiation is compatible with ChainRules backends that work in reverse mode, but not in forward mode?

[oxinabox]

The only fully ChainRules compatible forward-mode AD that is working right now is Diffractor.jl which is yet to tag a release. It does have a very nice forwards mode AD though.

ForwardDiff2.jl was abandoned.

ForwardDiff.jl works only via https://github.com/ThummeTo/ForwardDiffChainRules.jl which is opt-in per the rule.

[gdalle]

Yeah I did have Diffractor in mind, but also a potential future translator from ChainRules to EnzymeRules. I don't think the PR would be hard, I can draft it if there is interest

[ChrisRackauckas]

ForwardDiff2 was essentially replaced by TaylorDiff.jl which will be using ChainRules for its vector rules.

[oxinabox]

It would be good to add this

[gdalle]

It would be good to add this

Could this have something to do with a certain v0.2.0 release that just came out 😇 ?

[gdalle]

It would be good to add this

I can give it a shot but IIUC we can't test it on 1.9 since we need Diffractor?

[gdalle]

Also, the Diffractor backend already exists without being a special case of the (yet-to-be-written) ForwardRuleConfigBackend, is that an issue?