Update compatability with Flux 1.15 or 1.16

[camilodlt]

Hello and thanks for your package,

According to the flux releases, Flux 1.15 had two breaking changes :

The recurrent layers have been thoroughly revised. See below and read the [documentation](https://fluxml.ai/Flux.jl/v0.15/guide/models/recurrence/) for details.
    Flux now defines and exports its own gradient function. Consequently, using gradient in an unqualified manner (e.g., after using Flux, Zygote) could result in an ambiguity error.

and Flux 1.16 had one :

 The recurrent cells RNNCell, LSTMCell, and GRUCell forward has been changed to

and now returns the final output of the recurrent layer cell.

Are these changes preventing you from updating the compat ? I think 1.15 had some great speed/memory updates.

Another change that is not breaking yet is the deprecation of Flux.params => Flux.trainables

I could dev your package and tell you if your tests work or don't work with 1.15 and 1.16 if you want :)

[nic-barbara]

Hey @camilodlt, thanks for reaching out. Please feel free to dev the package and check if it works with later versions of Flux. Happy to review any pull requests required to migrate it to more modern versions.

At the moment, our focus is on our new Python version of the package, written in JAX (https://github.com/acfr/RobustNeuralNetworks), so if there are any major changes required to the package during the updates, I probably won't have time to implement them myself in the near future.