generate_gdd20_baseline needs SystemTest

[samsrabin]

It turns out generate_gdd20_baseline has been broken for a while, and I didn't notice because it wasn't being tested. Ideally it would have complete Python unit and system testing, but at the very least we should have a SystemTest in the ctsm_sci and crop_calendars suites like RXCROPMATURITY:

1. Do a multi-year run to generate outputs.
2. Call generate_gdd20_baseline on the outputs, generating new stream_fldFileName_gdd20_baseline.
3. Run at least a few days with the new stream_fldFileName_gdd20_baseline.

[wwieder]

Is there an issue with how gdd20 is currently being handled in the code (and in #3543), or is this just a testing issue to prevent future issues?

[samsrabin]

The latter: gdd20 in the FORTRAN code is fine as far as I know; #3543 is just to fix the Python script.

[wwieder]

requires 2 years of simulation, could be low resolution (f10)