This is a request for feedback to address #12 and #14 as it technically has some breaking api changes and I'm unsure what kinds of guarantees you have with downstream users.

New Features

• Bunch(**correctly_formatted_provider_dict) now makes a providers object from whatever's in the dict.
• providers[new_provider_name] = new_provider now works whenever new_provider_name is a str and new_provider is a Bunch, a TileProvider or a correctly formatted dict representing a Bunch or TileProvider
• providers.update(**{provider_name: new_provider, ...}) Now works wherever provider_name and new_provider match the __setitem__ criteria.
• Bunch.from_json(json_string) now does the work of the _load_json function. We could reasonably make this also load json files or ditch the method entirely and have providers.py make the json.loads call.

Breaking changes

• __init__, __setitem__ and update no longer function like their dict counterparts for the Bunch class. I'd be suprised if anyone depended on this functionality though.
• isinstance(TileProvider, Bunch) is no longer true, and TileProvider no longer has access to Bunch methods, consequently. Instead, TileProvider directly inherits from dict and implements it's own copy of __getattr__. I think this is technically more in line with what you want, but may have consequences I can't see.

Other thoughts

If you can provide more guidance about downstream use cases, I think it would be worth making the following adjustments:

• Changing the base type of Bunch from dict to collections.UserDict[str, Bunch | TileProvider] and adding in appropriate runtime validation to ensure it stays consistent
• Changing the base type of TileProvider to either a pydantic.BaseModel or a UserDict[str, str | int] and providing validation for required and optional fields.

I think both changes would provide better usability (ie would fail fast and noisily) if downstream users are adding their own sources and formatting them poorly.