Improve editorconfig options caching on ProjectState

[tmat]

Fixes #41840 by adding a new compiler API:

public class AnalyzerConfigOptions
{
    ...
    /// <summary>
    /// Enumerates keys of all available options in no specific order.
    /// </summary>
    /// <exception cref="NotImplementedException">Not implemented by the derived type.</exception>
    public virtual IEnumerable<string> Keys
         => throw new NotImplementedException();
    ...
}

Use this API in Code Style and Workspace layers to parse NamingStylePreferences from the options.

ProjectState maintains a cache of editorconfig options for given source file path. The cache holds on raw options in form of ImmutableDictionary<string, string> as returned by the compiler editorconfig parser. We currently wrap that dictionary in a new instance of DictionaryAnalyzerConfigOptions every time options are queried for the path. Elsewhere in the code we also maintain conditional weak table of the raw options dictionary to parsed NameStylePreferences for caching purposes. As a TODO in the code says this caching logic is rather fragile. In yet another place in the code we have reflection code that relies on the fact that DictionaryAnalyzerConfigOptions, and not any other type, is used to represent the options. This is also fragile since there is no direct relationship between the helper that uses reflection and DictionaryAnalyzerConfigOptions type.

This change introduces StructuredAnalyzerConfigOptions type that is used to cache both raw and parsed editorconfig options on ProjectState. The type itself provides facilities to cache NameStylePreferences (or possibly other expensive to parse options in future) as well as translation between Workspace, Compiler and Code Style layer representations of the options.

[tmat]

@333fred PTAL, compiler public API change.

[tmat]

@JoeRobich @sharwell @jmarolf @mavasani PTAL

[chsienki]

Compiler side changes LGTM, assuming we ok the overall API.

[333fred]

Compiler changes LGTM. The API needs to be reviewed by API design, but as long as this doesn't ship publicly before then it's ok to proceed.

[sharwell]

Review started but giving early feedback to address the comparer validation

[sharwell]

💭 This seems like a good candidate for ValueTask<>

[tmat]

Will follow up.