Is your feature request related to a problem? Please describe.
The MixedDensityEstimator currently takes a 1D discrete net and a continuous net and combines them autoregressively. To generalize this to multiple discrete dimensions, we should generalize this to an AutoregressiveMixedDensityEstimator which takes a list of density estimators. E.g. sth like

discrete_net1 = CategoricalDensityEstimator(condition_dim=x.shape[1])
discrete_net2 = CategoricalDensityEstimator(condition_dim=x.shape[1] + 1)
continuous_net = NFlowsFlow(input_dim=5, condition_dim=x.shape[1] + 2)

density_estimator = AutoregressiveMixedDensityEstimator([
    discrete_net1, discrete_net2, continuous_net
])