Refactoring data read

[Peter9192]

This reformats the dadta reading functionality. It includes the following changes:

• Formatting & remove unused imports
• Instead of one big function, make on function per model
• Use pathlib as it works on all operating systems
• Reduce duplication in reading ensembles

[Peter9192]

We should probably document/specify in attributes/units what exactly the variables in the resulting datasets represent.