robust_record refactor

[ceholden]

Refactor yatsm.algorithms.postprocess.robust_record to be more general. This function can be simplified to be refit(yatsm, prefix, predictor) and this would allow other regression methods (e.g., a cross-validated Lasso) to be ran within the same context of redefining the coef and rmse within each segment of each ensemble model.

Shortcut robust_record as refit(yatsm, prefix='robust', predictor=rlm.RLM).

Also refactor yatsm map --robust option to --result_prefix so we can output coef/prediction maps from robust_coef, lasso_coef, theilsen_coef, etc...

Config file specification:

YATSM:
    ...
    refit_prefix: robust
    refit_prediction: RLM
    ...

where RLM is a pickled RLM prediction object.

Merit: This should allow change to be fit using one method of prediction while also summarizing timeseries using other prediction methods.

[ceholden]

Should be good to close but waiting on some anecdotal evidence of it not messing everything up since test coverage is bad.