Update class modifiers specification.

[lrhn]

• Tighten the modifiers implied on anonymous mixin class applications.
• Claim that enum declarations are final.
• Allow pre-feature libraries to mix in platform no-mixin classes if they could before.
• A little rephrasing, to allow reusing a concept.

But do not allow interface mixin class etc. anyway.

@munificent @kallentu

[eernstg]

LGTM! A couple of comments added.

[leafpetersen]

For the record, I continue to think this is a mistake, for all the reasons laid out in #2723 .

[leafpetersen]

I am very much not on board with this change. If we actually feel that we need to ability to have variants of these that can be mixed in internally but not externally, we should instead remove the restriction that you cannot mix in a class which is not marked as mixin inside of your own library.

[lrhn]

Reverted the change to allow interface mixin class.

PTAL

[leafpetersen]

This looks great, thank you!

[eernstg]

I Indicated a handful of typo level issues (syntax errors in examples and such).

More importantly, the definition of 'prohibits inheritance' talks about 'declarations', and the actual wording of subsequent lines does not make sense unless we count each anonymous mixin application as a 'declaration' and consider that declaration to be 'marked with' each of its implicitly induced modifiers.

However, I still think the rules about anonymous mixin applications are simpler to understand if they are expressed directly in terms of the class modifiers (sealed as well as final/base/interface/none, explicitly specified as well as implicitly induced), so I inserted a 'suggestion' about how that could be done.

[eernstg]

It seems worthwhile to show that a mixin on S is usable, and to mention that the resulting class also cannot be implemented locally.

Suggested change

	base mixin MO on S {}
	base mixin MO on S {}
	final class DMO extends DM with MO {}

[eernstg]

Couple of comments.