fix(typing): Revise generated annotation order

[dangotbanned]

Description

The doc stated that None would appear after builtins.

altair/tools/schemapi/utils.py

Lines 855 to 864 in 2ba89d5

	def sort_type_reprs(tps: Iterable[str], /) -> list[str]:
	"""
	Shorter types are usually the more relevant ones, e.g. `str` instead of `SchemaBase`.
	We use `set`_ for unique elements, but the lack of ordering requires additional sorts:
	- If types have same length names, order would still be non-deterministic
	- Hence, we sort as well by type name as a tie-breaker, see `sort-stability`_.
	- Using ``str.lower`` gives priority to `builtins`_ over ``None``.
	- Lowest priority is given to generated aliases from ``TypeAliasTracer``.
	- These are purely to improve autocompletion

However, this assertion doesn't hold for any builtin with a name length greater than 4.

The issue is not new, but became more apparent to me while writing #3653 (comment)

Example

Below float appears after None

Before

altair/altair/vegalite/v5/schema/_config.py

Line 5004 in 2ba89d5

tooltip: str | bool | None | float | TooltipContentKwds

After

altair/altair/vegalite/v5/schema/_config.py

Line 5004 in e656661

tooltip: str | bool | float | TooltipContentKwds | None

Note

I was unable to work out a sorting strategy to achieve what the doc had intended.

Instead, I've opted to always place None last in the context it appears.
This is much easier to do, and is generally what most people would write (and expect to see).

In the non-generated code, this is already the pattern we use:

altair/altair/vegalite/v5/api.py

Line 2091 in 2ba89d5

format: Literal["json", "html", "png", "svg", "pdf"] | None = None,

altair/altair/vegalite/v5/api.py

Line 3494 in 2ba89d5

def _repr_mimebundle_(self, *args, **kwds) -> MimeBundleType | None: # type:ignore[return] # noqa: ANN002, ANN003

altair/altair/vegalite/v5/api.py

Line 3859 in 2ba89d5

self, row_limit: int | None = None, exclude: Iterable[str] | None = None

Preventing regressions

I've started small, but have included a few doctests (f9f0a53) that are now run alongside (20d5c17) those in altair, tests.

[dangotbanned]

Pinging @binste so that you see this when you're available again.

Happy to revert later if you have any objections 👍