Fix problem with factories in iris.util.new_axis().

[pp-mo]

It was found that adding an axis to a cube with an aux-factory had a subtle bug, which then causes errors when you copy the result :

Example:

>>> cube
<iris 'Cube' of mass_fraction_of_ozone_in_air / (kg kg-1) (model_level_number: 63; grid_latitude: 182; grid_longitude: 146)>
>>> 
>>> # Presumably, something here is wrong ...
... cube2 = iutil.new_axis(cube, 'time')
>>> cube2
<iris 'Cube' of mass_fraction_of_ozone_in_air / (kg kg-1) (time: 1; model_level_number: 63; grid_latitude: 182; grid_longitude: 146)>
>>> 
>>> # ... but this is what actually fails :
... cube3 = cube2.copy()
Traceback (most recent call last):
  File "<stdin>", line 2, in <module>
  File "/opt/scitools/environments/experimental/2019_01_31-1/lib/python3.6/site-packages/iris/cube.py", line 2967, in copy
    cube = self._deepcopy(memo, data=data)
  File "/opt/scitools/environments/experimental/2019_01_31-1/lib/python3.6/site-packages/iris/cube.py", line 3003, in _deepcopy
    new_cube.add_aux_factory(factory.updated(coord_mapping))
  File "/opt/scitools/environments/experimental/2019_01_31-1/lib/python3.6/site-packages/iris/aux_factory.py", line 156, in updated
    coord = new_coord_mapping[id(coord)]
KeyError: 140318647616680
>>> 

The new addition to the test code should demonstrate this (fails when run against old code).

[pp-mo]

What was actually wrong is :

• the dependencies within a factory must be the actual coords in the cube for cube.copy() to work..
• .. because otherwise the factory dependencies don't show up in the old-new 'coord_mapping' that the copy operation constructs
• .. whereas, "new_axis" returns a cube whose factory has its own independent copies of the original coords : so that won't copy properly

[lbdreyer]

I assume you'll update this with the issue number?

[pp-mo]

Ok

[lbdreyer]

This looks good 👍 ! Thanks @pp-mo!

Just need to update the comment with the correct issue/Pr number

[pp-mo]

Thanks @lbdreyer !
I will fix that shortly.
But I want to just hold off a little on this one, as I have since been made aware of a possibly-related problem in some user code, which seems to have a similar error ...

[pp-mo]

problem in some user code, which seems to have a similar error

That was not related after all.

Let's go ahead with this !
-- Except, I will retarget against v2.2.x
now re-targetted against v2.2.x

[pp-mo]

Following offline discussion with @lbdreyer :
Bugfix release whatsnew contributions should go in a separate block in the minor-version whatsnew documentation.
So here, moved note from a contributions file into a 'Bugs fixed in v2.2.1' section in the existing v2.2 whatsnew.
This parallels what we see in, for example, the v1.7 whatsnew bugfixes

[lbdreyer]

Great, thanks @pp-mo !