improve predict_DO output for pi, oipi, and sim models

[aappling-usgs]

see also #294 (for plots specifically): DO predictions for non-oi-only models require different methods from oi-only models.

• predict_DO.metab_bayes should make use of err_proc_iid, DO_mod, etc. if these are tracked and stored; otherwise they should give a message or warning describing how close we can get, or not, with the available information
• predict_DO.metab_model and mm_predict_DO_1ply use err.obs=err.proc=0 right now. they should use errors determined during the model fit, shouldn't they?