Example: BayesFlow Likelihood as serialized JAX function

Nobody has claimed this yet.

BayesFlow uses Keras throughout, which has become backend-agnostic in recent years.

Two key aspects for HSSM:

- Keras supports serialization of models (save/load)
- We should be able to use a Keras model as a simple JAX function finally

To close this issue, produce a simple example of inter-operability between BayesFlow and HSSM.

1. Use `ssm-simulators` to instantiate a basic DDM simulators
2. Train a likelihood networks via BayesFlow
3. Save the model
4. Load this model and turn it into a differentiable likelihood that works with the HSSM low-level interface
5. Translate the use case to the high-level interface

The output here should be one or two Jupyter notebooks that can be part of the docs as initial examples.

Start by reviewing the existing HSSM documentation and the low-level and high-level interfaces, then examine how ssm-simulators and BayesFlow are used together. Build one or two Jupyter notebooks covering DDM simulation, likelihood-network training, model save/load, differentiable JAX conversion, and both interfaces. Done means the notebooks run as documentation examples and demonstrate the complete interoperability workflow.

Written by the indexing model from the issue text.