See #122713 for context.

We should investigate how difficult it would be for us to convert the implementation to building directly on ffi (as we do for path_provider_linux already). While win32 is convenient, it adds external exposure to one of our most foundational packages, which is not ideal (see also #81222).

If it's just function boilerplate, then we should probably just re-implement it. If there's more complexity in what we use, we can evaluate the cost of re-implementing it.