Fix copy to include Ancillary variables and Cell measures

[stephenworsley]

Fixes another issue in #3483.

[stephenworsley]

This should be merged after rebasing with #3536 so that equality works for testing.

[stephenworsley]

This has now been rebased with #3536.

[trexfeathers]

One small change, and please confirm whether this is still "DO NOT MERGE" since the rebasing has been done? Thanks @stephenworsley 😊

[trexfeathers]

While AV's are a new feature, CM's are not. So I think there should be a what's new bugfix entry for the fact that CM's are now copied

[stephenworsley]

I have patched in a rough proof of concept fix for equality. This ought to be improvable once #3551 is merged.

[trexfeathers]

Notes on the effects on all uses of cube.copy. See below for a full list of uses. For those where a problem may be caused I have created #3554, as the problems are relatively minor and can be handled separately, so that this more important fix can proceed now.

• analysis.PercentileAggregator.post_process ✔
  • used in collapsed, which is also being modified to retain AV's in PI-3483: Ensure collapsed/aggregated_by/rolling_window treat ancillary variables properly #3549 so these should be released together
  • allows the end result of a percentile aggregation collapse to retain AV's, CM's
  • for this there are no considerations needed about retention/discarding depending on the collapsed dimension
• analysis.maths._math_op_common ✔
  • allows AV's and CM's to be retained after performing arithmetic on a cube - the first cube if multiple cubes are involved
  • no foreseen issues
• analysis.calculus.cube_delta ✔
  • actually needs this copy fix now that __getitem__ has been fixed in __getitem__ gets ancillary variables #3548 - these should both be fixed together
  • (otherwise behaviour is different is coord(s) are circular or not)
• analysis.calculus._curl_subtract ✔
  • used in the production of curl cubes
  • original dimensions are retained in the output curl cubes, so no unusual consequences expected if the output cubes now include AV's/CM's
• analysis._regrid.CurvilinearRegridder ❌
  • see Ancillary Variables - validate 'all other' cube operations #3554
• analysis.trajectory.UnstructuredNearestNeigbourRegridder, also .__call__ method ❌
  • see Ancillary Variables - validate 'all other' cube operations #3554
• analysis._grid_angles.rotate_grid_vectors ✔
  • original dimensions are retained in the output cubes, so no unusual consequences expected in the output cubes now include AV's/CM's
• analysis._grid_angles.rotate_winds ✔
  • see points on analysis._grid_angles.rotate_grid_vectors
• experimental.stratify.relevel ❌
  • used for interpolation over a dimension denoted with levels
  • see Ancillary Variables - validate 'all other' cube operations #3554

[trexfeathers]

Thanks @stephenworsley, LGTM ✔

[pp-mo]

This is a public function, so you really need to extend the docstring to cover the changes.
The new 'object_get' keyword is undocumented.
It might also be desirable to rename some things here, since e.g. "coords" is no longer always what it says.

Another way is to create an enhanced, private inner routine + leave the public one as a thin wrapper for coords alone.
Given the naming of the function itself, that might actually be preferable.

[stephenworsley]

I've opened #3558 to address this.