Native autograd interface

[yebai]

No description provided.

[gdalle]

Note that @wsmoses and myself added comments on the now deleted commit 7972f1a, feel free to take them into account

[yebai]

@gdalle, given your limited familiarity with the Turing.jl codebase, it may be best to add your comments only where explicitly invited.

[gdalle]

I did not comment on the Turing-specific parts, but I thought you might want to benefit from my familiarity with DifferentiationInterface and all of its backends, and from Billy's familiarity with Enzyme. Voluntary reviews are usually welcome in open-source projects

[yebai]

Your enthusiasm and engagement are noted and appreciated. I don't think we will benefit more from comments on DI at this stage -- your comments created a lot of attention/pressure on whether DI should be a hard dependency of Turing.jl, which is beyond feedback.