MVS with option na.action = "pass" trains a meta-learner

Current behavior: MVS with option na.action = "pass" trains a meta-learner on the complete subset of the data (i.e., it trains a meta-learner on na.omit(Z)). Note that this does not affect the rest of the results in any way.

Desired behavior: MVS with option na.action = "pass" should skip training the meta-learner entirely.