Review and improve rolling data and coordinate points for aggregation by hour_of_day

@jfrost-mo is already working on this.

Since Jun 17, 2025.

In #1450, the following technical review comment and suggestion was raised in relation to method to ensure time/hour coordinate and the data array are sorted in time order.

====

I'm going to suggest a different approach to sorting the cube. The current method works, but has two major flaws.

1. It relies on the layout of data array within the cube remaining consistent. While this may be the case for UM/LFRic data, we can't rely on it in general.
2. The use of the `.data` cube attribute causes the data to be realised. Currently there are a few places in CSET where this will happen, but it is something we are going to have to get rid of if we want to support high resolution or long duration data sets, such as K-scale.

Instead of dealing with the underlying data array, we can blow the cube apart and construct a new one, with the relevant coordinate sorted. Some code to do this is below.

```python
def sort_cube_by_coordinate(cube: iris.cube.Cube, coordinate: str) -> iris.cube.Cube:
"""Sort a cube by one of its coordinates.

The cube is broken apart and reassembled to sort the specified coordinate.

Arguments
---------
cube: iris.cube.Cube
Cube to sort.
coordinate: str
Name of coordinate to sort. It must exist on the cube.

Returns
-------
iris.cube.Cube
Sorted version of the input cube.
"""
slices = cube.slices_over(coordinate)
# Sort slices by the (scalar) points of the coordinate.
sorted_slices = sorted(slices, key=lambda c: c.coord(coordinate).points)
return iris.cube.CubeList(sorted_slices).merge_cube()

by_hour = sort_cube_by_coordinate(by_hour, "hour")
```

_Originally posted by @jfrost-mo in https://github.com/MetOffice/CSET/pull/1450#discussion_r2152893658_