`update()` function breaks when all alternative models for a moderator gets assigned zero prior probability

[FBartos]

The issues is in .ensemble_inference() function when the computing BayesTools::marginal_inference()

• information about the factor levels from models with only null prior is missing

[FBartos]

unsuccessful trials

• remove all null predictors (clears the way up to the marginal inference)
• then cleaning up the formula produces further issues

potential solution

• adding factors levels information before forwarding to BayesTools
• allowing BayesTools to take the factor levels information from spike priors (if specified as well)