Several more fields on ModelElement should be eagerly computed

[srawlins]

In PackageGraph._precacheLocalDocs, we iterate over _allModelElements and ask each element:

• .isCanonical, then, if true,
• .canonicalModelElement == null , then, if true,
• .enclosingElement!.isCanonical.

ModelElement.isCanonical calls:

• .isPublic,
• .canonicalLibrary, and more.

ModelElement.isPublic will check for some non-public validations (name is not empty, in a public library, enclosing element is public), and if we pass all of those, we get to !hasNodoc, which requires parsing the doc comment.

So many of these data should be calculated eagerly, as a ModelElement is being constructed. In particular, isPublic and isCanonical. For public elements, this means calculating hasNodoc eagerly, but we probably still want to avoid parsing doc comments on non-public elements.

Doing all of this eagerly should turn some late final fields into final and simplify the code.

Similar to #2970, in that we may compress the doc comment parsing in the process.

[jcollins-g]

This turns out to be difficult as the mentioned fields indirectly depend on recursively constructing ModelElements for other elements that, if you precompute these values, could result in a construction loop.