improve plot_DO_preds for pi, oipi, and sim models

for `y_var %in% c('conc', 'pctsat')`, `pi` models should have prediction line segments that each start at a DO.obs value and progress toward the next one. what we have now is what might be called the DO.pure predictions - what we get if we extrapolate the daily metab params to DO.mod assuming no error. `oipi` models should have the state-space estimates in `DO.mod`, which are pi-corrected but don't need to be line segments starting at DO.obs values.

with bayesian models we can track inst estimates of err_obs_iid and err_proc_iid, which could help with both pi and oipi plotting.