[3.0] Cache ClangSharp outputs & use cache if we shouldn't run ClangSharp

[Perksey]

This introduces a caching mechanism in SilkTouch 3.0 so that we can cache the raw outputs from ClangSharp to allow the rest of the SilkTouch pipeline to be run & developed on platforms other than the ones in which ClangSharp can run. For example, we'd cache raw outputs for the Windows SDK bindings so that if we don't have access to the Windows SDK for example (e.g. we're on macOS), we'd open the cache and feed the cached outputs through the pipeline.

The way I envision this looking is:

• The bindings that can't be regenerated outside of a specific environment are cached upon regeneration in that environment.
• We'd have a CI job that will run in that specific environment to regenerate the cache when there's an update to the underlying dependency, similar to how dependabot works.
• This keeps all developers regardless of platform to be able to clone Silk.NET and regenerate all the bindings, cached or uncached as appropriate.

This was also how the BuildTools cache was intended to work, but I don't know of a single instance of it working properly and even if it did, the cache was output at a pretty late stage in the pipeline such that the only changes observable without regenerating the cache were pretty much just overloads.

[Perksey]

Oh and this also lets us use Git repos & NuGet packages as ClangSharp inputs, alleviating the need for submodules which have to be manually synchronised.

[github-actions]

File	Coverage
All files	10%	✅

Minimum allowed coverage is 0%

Generated by 🐒 cobertura-action against 89dcc7a

[Perksey]

I think this is ready for review, I'll do the NUKE/CI stuff in a later PR.

The LOC count may seem excessive, but this is because TerraFX has updated and as a result so has our Windows bindings.

The bit I'm interested in scrutiny for is the stuff in sources/SilkTouch as this has been written over many sleep-deprived evenings.

[curin]

Looks good from what I can tell. Didn't really look at the d3d12 code, since I'm assuming that is just raw clangsharp output atm. The actual code for Caching, Sources, and the new AddIncludes looks all pretty sound to me.
The generated OpenGL code also looks a bit better formed.

[Perksey]

Thank you all for reviewing and your continued support with the 3.0 effort!