Using `JuliaFunction.apply()` from Python only takes one tensor input argument

[ShantMahserejian]

When calling JuliaFunction.apply() from Python, and when trying to pass multiple input arguments including tensors, we are limited to passing only a single tensor variable that needs to be packed/unpacked outside/inside the Julia function call being created. For example,

JuliaFunction.apply(f, x)
where x must be a single torch tensor.

Possible solutions for lazy python implementation can include:

1. Ideally, enable keyword arguments JuliaFunction.apply(f, *args, **kwargs) to allow multiple, labeled input arguments
2. At least, enable non-keyword arguments JuliaFunction.apply(f, *args) to allow multiple input arguments without having to pack/unpack

This is nice to have, but not blocking.

[ShantMahserejian]

@ChrisRackauckas: please add any clarifications that you may have

[LilithHafner]

We currently use ForwardDiff.jl for autograd, so we cannot differentiate with respect to multiple inputs due to https://juliadiff.org/ForwardDiff.jl/v0.10.2/user/limitations.html. I'll look into swapping out the backend to Enzyme.jl