more consistent treatment of min_nobs

[rfiorella]

version 0.6.4 introduced a minimum number of observations for the calibration regression - modified regression function and high-level wrapper (calibrate_carbon). However, after reviewing a bit more to update water functions, there's another observation threshold in the restructure* functions (e.g., L. 222 here).

Would also be good to move some of the other thresholds that are in the restructure and regression functions out to have access from the high-level wrapper for 0.7 as well.