Move VARIANT support to the windows crate

[kennykerr]

#2786 added richer support for VARIANT and PROPVARIANT and at the time the only practical way to do that was adding it to the windows-core crate. But that meant that everyone using the windows-core crate carried around VARIANT support even though that's really only meant for some very old Windows APIs in the windows crate. Now that #3110 added a simpler way to handle type extensions, we can more naturally provide this same functionality directly in the windows crate, unburdening the windows-core crate and simplifying the windows-metadata and windows-bindgen crates as well with less "special" type support.

[kennykerr]

It's tricky because VARIANT and PROPVARIANT are in different Win32 metadata namespaces. I might have to break the dependency apart and implement them separately.

[ChrisDenton]

This might be a dumb question but do they need to be in a separate namespace? Or is this something that could reasonably be changed?

[kennykerr]

Yes, that would greatly simplify this. Arguably they should just be together as the two are interdependent APIs.

[kennykerr]

It's far from perfect but greatly simplifies windows-core and windows-bindgen and lets me keep pursuing improvements to code generation so it's a step in the right direction.