Central model records

[qin-yu]

Design Changes

• Model zoo yaml record was checked unnecessarily each time get_model_zoo() or list_models() is called, now this zoo information is stored in a ModelZoo singleton.
• Model metadata were filtered or accessed inefficiently in a dictionary, now they are nicely organised in ModelZoo singleton's pandas.DataFrame object.
• In the end, /plantseg/utils.py only contains real utility functions, and zoo.py has classes with the model related functions.

Coding Changes

• Use pydantic v2 for config checking, starting from this new ModelZoo with ModelZooRecords
• Constant file names and paths are all moved to __init__.py and are all manipulated in pathlib.Path instead of by os. Relevant functions are all changed accordingly, with some refactored.

To-do

• [x] Add ModelZoo class
• [x] Use table-like internal representation
• [x] Check "get" methods
• [x] Check "add" methods
• [x] Check "set" methods
• [x] Check "del" methods
• [ ] Add tests

[qin-yu]

Hey @lorenzocerrone, I've developed a central record singleton class for managing models—a runtime table for storing information about models. This setup paves the way for integrating default models, custom models, and models from bioimage.io. Although bioimage.io is currently undergoing migration, the new database will soon be ready for integration.

Could you review this when you are fixing issue 218 today?

[qin-yu]

Replying to https://github.com/hci-unihd/plant-seg/pull/220#issuecomment-2051378622

Yes I'll use Pydantic for all necessary config validation. But I'll do it bit by bit like what I did for pathlib.