[Explore] Investigate the differences in xCDAT's annual cycle climatology compared to E3SM Diags climo.py/climo_xr.py and cdms2/cdutil

[tomvothecoder]

Discussed in #845

^{Originally posted by tomvothecoder September 4, 2024}
This discussion is an extension of PR #688 in xCDAT.

Notes:

• e3sm_diags/e3sm_diags/driver/utils/climo_xr.py

  Lines 132 to 134 in 9be8cea

  	# NOTE: Since `time_bnds`` are decoded, the arithmetic to produce
  	# `time_lengths` will result in the weighted averaging having an extremely
  	# small floating point difference (1e-16+) compared to `climo.py`.

• climo.py and climo_xr.py use masked arrays with masked averaging , while Xarray/xCDAT uses np.nan to represent missing values
• Weight generation is different in CDAT (casts float32 to float64) (source)