Codegen Improvements

[pavelkouril]

At first I just wanted to do small cleanup (e.g. utility methods to reduce repeating code etc.), but in the end I'm opting out for a bigger changes. My goal is to split the codegen from 2 logic passes (build + generate) all happening in the *Instance classes, into 3 more separate passes.

The result should be a less hacky and less error prone generator. 🤞

High-level overview of the changes, with description of the pass splits:

1. Parsing will parse the metal-cpp headers into the *Instance model classes and will also collect the selector definitions
2. Linking will take the parsed model data and resolve selectors to the actually parsed definitions, type hierarchy and others and basically prepare model representing the final C# code with C# namings
3. Codegen goes over the C# model from linking and just emits it with more generic codepaths

Naming is not final, I am open to suggestions.

Current state is nearly final, just need to cleanup the changes & finalize naming. Other than that, the changes fixed some wrong selectors and now the output is stable and does not need any manual touch-ups.

[pavelkouril]

Please review. I'll fix any coding standard violations or naming you dislike.

If you'd like to unify some of the codegen/csharp model APIs, I can do it as well (but I wanted to keep the csharp representation on the simpler side).

Running the codegen emits 0 changes and requires zero manual touchup, meaning it can be finally considered stable.

I split the commit into two:
1/ Codegen changes
2/ Re-run the codegen to fix any whitespace inconsistencies + issues with wrong selectors in previous releases

[IsaacMarovitz]

Lgtm apart from a few comment nitpicks. Thank you for tackling this!

[pavelkouril]

I fixed the comments and rebased + force pushed.