Fix bug by adding attribute self.epistemic_optimizer with learning_rate option

[jeipollack]

This PR closes #1 by adding the missing initialization of self.epistemic_optimizer in the DiffusionActiveInference class, which triggered an AttributeError during training.

The following line was added:

self.epistemic_optimizer = torch.optim.Adam(
    self.epistemic_estimator.parameters(),
    lr=self.config.epistemic_lr if hasattr(self.config, "epistemic_lr") else 1e-4
)

To test:
Code was launched and training has proceeded well over 4 hours without the error