Fix .metpy.units checking .data for units and loading data into memory

[dcamron]

Description Of Changes

• Removes check_axis check in no gridmapping fallback in parse_cf (unnecessary via Remove misleading fallback to lat/lon CRS when only 2D lat/lon present #1651)
• Points .metpy.units to xarray.Variable._data instead of .data in checking for quantity with units to avoid loading data into memory
• Test to see if .metpy.units touches .data. If there are any thoughts on more thorough tests for this situation let me know

Checklist

• Closes parse_cf loads all data into memory when grid_mapping is missing since .metpy.units checks .data #1748
  • Not sure if we want to leave open for larger discussion
• Tests added
• Fully documented

[jthielen]

Looks good to me! I say we either leave #1748 open to track progress on the more permanent fix (once there is a public API for the type check) or open a new issue...just don't want to have it forgotten in a closed issue.

[dopplershift]

#1835 can track our use of ._data.