Missing Average Energy for Mixture node

[abpolym]

I don't know if it is defined somewhere else, but the Mixture node does not have an @average_energy definition

[bvdmitri]

tagging @bartvanerp here

[bartvanerp]

True as the mixture node operates with scale factors only and effectively performs model comparison. Hence, I am not sure whether computing BFE in cases where this node is used makes sense at all. However, might be good still define it and throw a warning or something.

P.S. for Gates (https://papers.nips.cc/paper_files/paper/2008/hash/4b0a59ddf11c58e7446c9df0da541a84-Abstract.html) a BFE could be defined

[abpolym]

Alright let me know if you like this solution #546 @bartvanerp

I will add tests for that average energy as soon as my tests in my other PR #539 are merged

Maybe one follow-up question:
Could we compute an expected energy contribution depending on the value/type of q_switch? So if q_switch is one-hot, we return the component's average energy, if q_switch is "soft" we average the component energies and add log q_switch(k) or something?
I am not sure if it makes sense and it is probably an approximation, but it is better than 0.0 in the current proposed solution?