Remove all unwanted cube attributes in read.py

### Something else
A few of our models contain unwanted attributes (like cube history, date of processing), which prevent merging of multiple cubes into a single cube (as the attributes differ). A fair number of attributes are useful (such as LFRic `time_origin` which is used to reconstruct `forecast_reference_time` and `forecast_period`), and `model_name`.

We could remove all attributes that don't match with `iris.util.equalise_attributes`, though we might end up removing useful attributes. We should have a list of attributes we want to preserve, and remove the rest.