Teach GenApi to understand nullability

[safern]

I marked the PR as WIP since I still want to implement a cache for the MetadataReader, use the document pointers used by the HostEnvironment to load the assemblies instead of using the locations and opening a file and creating a PeReader every time. (However this is ready to use in the meantime).

This teaches GenAPI to understand nullable attributes and emit ? correctly.

While doing this work I found that GenAPI didn't understand DynamicAttribute and didn't support nested ValueTuples so I fixed those as part of this work as well.

I needed to pull in a dependency into System.Reflection.Metadata in order to be able to get the custom attributes from GenericConstraintTable and InterfaceImplTable since CCI doesn't expose the attributes for those guys, making it imposible to annotate interfaces, generic constraints and explicit implementations correctly. The reasoning of pulling in System.Reflection.Metadata instead of modifying CCI was because we want to move out from using CCI sometime soon.

** Here you can find a gist of a sample input and output after these changes **

cc: @dotnet/nullablefc @danmosemsft

[terrajobst]

Nice! The dependency on S.R.MD is unfortunate, but it is what it is.

Long term, I'd like us to move away from MS.Cci and move to S.R.MetadataLoadContext instead. However, this probably requires us to expose some these things in reflection as well.

[ericstj]

Even better might be to use Roslyn if possible.

[terrajobst]

Would you mind rebasing this on top of master again? This way, I can get a coherent build of API Reviewer that can diff attributes and see nullable annotations.

[safern]

Yes. I will rebase when I get back to it and update it to understand the new compact metadata.

[safern]

Pending:

• Add tests.
• Add MetadataReader cache.

[safern]

@ericstj I added the MetadataReader cache, and it improved the execution time by half a second so I think it is worth having it.

Also at the end I decided to not add it to the Host as I didn't want to make the CSDeclarationWriter dependent on the Host as it seems like they were intentionally decoupled, since the CSDeclarationWriter only needs an IEnumerable.

[safern]

@ericstj this is also ready for review whenever you get a chance.

[safern]

@dougbu this change will also need updated refs in the ingestion PR for asp.NET. I added an space after inlined attributes. Ran code check locally and that is the only impact of this change in aspnet extensions.