estimate_abundances() should fill in missing variables with default values

[ctrapnell]

The user might have included a bunch of technical or nuisance covariates in their model. When they want to do a contrast between the interesting variables, but neglect to include others, we should fill them in with reasonable defaults. For example, if they have a "batch" term, we could add a "batch" column to their newdata dataframe prior to calling predict(), filling in a default value for their row. We should emit a message to this effect.