[peewee] add typed fields, queries, and update to some of the new APIs

[coleifer]

Attempt to get Peewee's typing a little bit nicer and bring in some of the additions since 4.0.5.

Added:

• JSONField + associated methods
• Make field access on model class vs model instance work (see note)
• Also get Field init params working properly for field-specific kwargs
• Add typing to query results
• Add typing to a handful of the chainable methods

For field access - peewee uses metaclass magic along with descriptors, kinda like Django. Fields are now generic over their Python value, so Model.field is the field instance for building queries, and instance.field is the value (e.g. str for CharField, related model for a ForeignKeyField, etc). Also specifying null=True makes the instance value optional, which is nice.

[coleifer]

Alrighty, I think this one's ready for review any time. Let me know if you have any q's or feedback - I think these changes will massively improve ergonomics for peewee users (all 3 of us lol).

[srittau]

A general comment: Non-obvious Anys need to be documented, i.e. what are the acceptable types and/or why is Any used here. Obvious Anys include dict[str, Any], Callable[..., Any], or Field[Any]. But for example, in _FieldKwargs default: Any and choices: Iterable[tuple[Any, str]] | None need to be explained with a (short) comment, for example:

    default: Any  # Accepted types depend on the field type

(I'm not sure whether that's the correct explanation.)

[github-actions]

According to mypy_primer, this change has no effect on the checked open source code. 🤖🎉

[coleifer]

@srittau - Those should be addressed now in 8413f13. Were there others you were concerned about specifically?

[coleifer]

@srittau / team - it really would be fantastic to hear what remaining blockers there are from your side. These changes, I believe, will seriously improve the utility of the stub for Peewee users. Like Django and others, we have a really dynamic API and use metaclass magic and descriptors for field access. These changes address differentiating ModelClass.field vs model_inst.field, something that's historically been a pain point, among a number of other improvements. I hope you will consider merging.

The distribution channel for a second/separate set of stubs is just not there, IMO, so I'm really hoping that these changes can get upstreamed. I can go into more detail about this if you wish, but I think the best value and least pain for existing users is typeshed.