Improve type checks

[larsoner]

1. Move from mypy to ty, including as pre-commit as ty is fast enough 🚀
2. Add a consistency check for documented types (NumpyDoc) and our types (this check could eventually move to docstub but let's use it here for a bit)
3. Stricter enforcement (mypy missed some stuff) and checking in epochs.py, evoked.py, and mne/io
4. Correctly type the io.read_raw_*, but keep their docs/ / website aliases pointing to publicly documented objects
5. Most typing changes were of the form 1) dict[str, Any], 2) assert x is not None for when x is self._data etc. (places where we know it is by code structure), plus using __getattr__ and __setattr__ to get *Bunch classes to play nicely
6. Some simplification of ctf/res4.py because type checkers have trouble with NumPy structured arrays
7. Various minor fixes caught by the tests + checker, like instance of RawEDF where we should have had instance of RawBDF

Used Claude Opus 4.8 to help me with the drudgery of assert x is not None etc., but I also made a lot of changes myself, and understand + vouch for all code.

[larsoner]

Okay too many packages needed for type checking made pre-commit.ci unhappy, so here I move to prek + prek-action in autofix.yml instead. pre-commit.ci will be red here, but once we're happy with this PR we can merge it and disable pre-commit.ci and use prek. (prek uses the preexisting .pre-commit-config.yml and is a nice, faster version of it.) We could also consider other toolsk, but prek is the least invasive to existing dev workflows (people can continue to equvialently continue using pre-commit locally if they want!)

[larsoner]

A tiny bit of stuff here would get simpler if we dropped 3.10 support. 3.11 is actually almost out of SPEC0 so I don't mind dropping 3.10 now. But not too painful to keep it. It's EOL in October.