[FEATURE] Support Mixture of Gaussians (and other parametric distributions)

[han-ol]

Is your feature request related to a problem? Please describe.
Other

Describe the solution you'd like
A MixtureDistributionScore that is inheriting from ParametricDistributionScore and gets constructed with a dictionary of ParametricDistributionScore s. Methods and attributes will need to be forwarded. head names can be f"{component_name}__{head_name}".

Describe alternatives you've considered
Separate inference_network class that does not utilize bayesflow.scores makes no sense.
Main alternative would be to NOT support mixtures.

Additional context
NeuralEstimators.jl implements this in: https://msainsburydale.github.io/NeuralEstimators.jl/dev/API/approximatedistributions/#NeuralEstimators.GaussianMixture
and I this might be where sbi gets them from: https://github.com/bayesiains/nflows/blob/master/nflows/distributions/mixture.py