[ENH] Saving models when testing deep learning estimators

[TonyBagnall]

Describe the feature or idea you want to propose

as I understand it, deep learning estimators save a model by default, as is standard practice in the field. This means that all our testing saves models, and this causes occasional CI fails (and is unnecessary really)

Describe your proposed solution

set default test parameters to not save a model, then introduce a specific test to test saving that uses a temp directory

Describe alternatives you've considered, if relevant

No response

Additional context

No response