Fixes to interpolation to pressure levels

[djgagne]

I made some significant revisions to the pressure level interpolation module to accommodate the issues @yingkaisha 's PR raised and to make the pressure interpolation process for geopotential and temperature consistent with CESM/ECMWF. New features:

• surface geopotential is now passed into credit.interp.full_state_pressure_interpolation as a numpy array rather than embedded in xarray Dataset with other state variables.
• output.py updated to reflect changed API.
• temperature and geopotential are extrapolated below ground based on Trenberth et al. 1993.
• mean sea level pressure reduction implemented based on Trenberth et al. 1993 and CAM fortran code. The formula adjusts the temperature and lapse rate correction based on the surface height. According to the paper you might still see some wonkiness in the Himalayas/Tibetan plateau, which would require some west-east interpolation over mountains.
• pressure to hybrid coordinate interpolation implemented. Interpolation routine ignores levels below ground but also requires surface pressure field to do so.

[djgagne]

Everything should be ready for review now. Test passes and inference code runs on casper.

[djgagne]

I have increased confidence that the IFS postprocessing documentation is wrong about two things. Equation 3 has ps and pL flipped, as does equation 6. It also looks like the height-dependent temperature corrections are being used to extrapolate geopotential after all. My answers are now really close to those from ERA5 and are probably about as close as I can get without direct access to the FULL POS source code, which I am still working on getting. @jsschreck and @yingkaisha please look over because I want to get everything merged this afternoon.

[djgagne]

Thanks John! Merging now.