Attempt to statically collect types in typeshed (ahead of time)

[lagru]

A first shot at #42.

This is an attempt to build a map of all usable type imports in Python's _typeshed. That would make docstub a lot more able to match doctypes to a valid Python type.

I initially thought, I had gotten pretty far by re-purposing the TypeCollector class. Unfortunately, _typeshed makes heavy use of re-importing types. E.g. the content of collections.abc:

from _collections_abc import *
from _collections_abc import __all__ as __all__

This means the current approach in tools/collect_stdlib_types.py will miss a lot of types.

This means resolving references and semantic analysis which is a lot more tricky to do. I think it's possible but would take a lot of work that's not actually the focus of docstub. It would also re-implement the wheel since mypy, basedpyright, etc. must have already figured that out.

So from that point, I looked into mypy's internals (including stubtest & stubgen), aiming to leverage mypy to create the list of types in src/docstub/_stdlib_types.py. To be honest, I couldn't make a lot of sense of it, even armed with a debugger. 🙈

Besides asking for support on the appropriate mypy channel, I don't have a lot of concrete ideas on proceeding right now. I could explore using a proper language server. But that seems similarly complicated right now.

I could use intersphinx mappings like suggested in #22. But I'm not sure if that actually contains all types. It's an interesting approach. Not all third-party packages may provide an intersphinx mapping but there might be packages that do but don't have stubs available. And it's already pretty close to what I actually want as output...