Add aggregation CLI

[btrotta-bom]

Add CLI for iris aggregations sum, mean, median, std_dev, min, max. The code wraps the existing function utilities.cube_manipulation.collapse_realizations, adding the option to specify different aggregation methods (with "mean" as the default, so that existing code will continue to work).

Testing:

• Ran tests and they passed OK
• Added new tests for the new feature(s)
• Added acceptance tests

[codecov]

Codecov Report

All modified lines are covered by tests ✅

Comparison is base (7e2ed68) 98.39% compared to head (36fe9ed) 98.39%.
Report is 1 commits behind head on master.

Additional details and impacted files

@@           Coverage Diff           @@
##           master    #1952   +/-   ##
=======================================
  Coverage   98.39%   98.39%           
=======================================
  Files         123      123           
  Lines       11808    11814    +6     
=======================================
+ Hits        11618    11624    +6     
  Misses        190      190           

Files	Coverage Δ
improver/utilities/cube_manipulation.py	99.21% <100.00%> (+0.01%)	⬆️

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[benowen-bom]

This PR adds functionality to enable aggregation of specified dimension coordinates by suitably wrapping the collapsed function contained utilities/cube_manipulation.py.

The code has been implemented in a general fashion so as to allow aggregation of any dimension(s), as well as broadcasting back to the original dimension coordinates. However, there are some considerations around threshold coordinate based data and the metadata for broadcasted datasets that need further consideration.

Hopefully the changes to address these issues shouldn't require too much work. In the case of the threshold dimension, I'm not advocating that quantities should be evaluated via integration of the probability distribution, rather that the quantities evaluated using the current approach are not the analogues of those evaluated using "realization" or "percentile" based data.

[btrotta-bom]

@benowen-bom Thanks for pointing out that Improver already does broadcasting. This allows us to simplify this change considerably. The code now just slightly extends the existing collapse_realizations method to allow the user to specify the aggregation type. (We currently do not need to aggregate over any other dimensions, so I've left this functionality out.) The option to rename the cube is now handled in the cli.

[benowen-bom]

Thanks for making the updates from the previous review. The changes made have simplified the usage cases significantly and avoid the problematic cases I raised previously. I've got a few suggestions, but these are relatively minor.

Re unit-tests, given no new tests are added (at the module level), I think it's fine to use unittest rather than pytest.

[benowen-bom]

Thanks for making those final changes. This PR looks good to me and provides a means to evaluate aggregate values from realization data, as needed for rainforests calibration.

I think rolling back the generalised elements was a good call. I think there is value in these, but they can be added when the need arises and we have the time to think through them carefully, particularly around threshold based data.

[gavinevans]

Thanks @btrotta-bom 👍

I've added some minor comments.

[gavinevans]

Thanks for the updates, @btrotta-bom 👍