add a prediction function and method to handle forward simulations

[LegrandNico]

Now that #274 is merged, we need to implement the corresponding scan function so the main Network class can sample predictions when provided with time step sequences.

The main call should look like this:

predictions_trajectories = Network.predict(n_predictions=5, time_steps=np.ones(20))

Where predictions_trajectories should be a dictionary of arrays or size (n_predictions, time_steps) each (i.e. just like node_trajectories but for multiple simulations).

The predict method should handle the creation of the scan function, and the vectorisation of random keys.

[LegrandNico]

@SylvainEstebe