Handle and document multi-chain BCF with an internal propensity score

[andrewherren]

In the case where a user wants to run multiple independent BCF chains and hasn't provided a propensity score, we currently handle this by using the first chain's internal propensity model as the internal propensity model in the "combined" multi-chain model. Going forward, we should:

• Use the internal propensity model from a warm-started BCF model, if one is provided
• Document this in closer detail