Fine grained CSS support

[amolenaar]

PR Type

What kind of change does this PR introduce?

• Bug fix
• Feature
• Chore (refactoring, formatting, local variables, other cleanup)
• Documentation content changes

What is the current behavior?

CSS can only be applied to elements (class, association). We can not write CSS that target specific parts in the element, e.g. attributes shown in a Class.

Issue Number: #2919, #3023

What is the new behavior?

• I added a new shape type: CssNode. With this shape we can tell Gaphor to update the style sheet for a specific element, such as an attribute owned by a class.
• Add support for the ::after pseudo-element. This allows us to create custom tags (e.g. {abstract} to make it more clear a class/operation is abstract).
• Added more type annotations.
• Computed styles are cached
• Style names have been applied to names (at least the ones shown in bold) and compartments.

Does this PR introduce a breaking change?

• Yes
• No

Other information

There's still some work ahead:

• Add CssNodes for all elements that makes sense.
• Move inline styling to a style sheet.
• Support for classes, so we can address styling for compartments in a Class item as .compartment. I think this makes more sense that referring to it as an element.
• Currently the CSS model is rooted in the Element. E.i. an element that is contained by another element does not inherit any properties from its owner. Should this change too? (Make style sheets cascade #2977)
• Add a way to figure out what CSS nodes are used in an item. (Make style sheets cascade #2977)

!! NB. The caching is still a bit flaky. I disabled caching for now, but that may result in a lot more style computations.

[danyeaw]

This is a really nice feature!!

Would you mind please adding a changelog entry so that we keep the release ready?

[amolenaar]

@danyeaw I added a change log entry.

I'm not really happy with how pseudo styles and inherited styles are calculated (with the "private" -gaphor-style-node and -gaphor-compiled-style-sheet CSS properties). Instead it should be a proper class, I suppose. It would be really nice if we can keep the type information TypedDict is providing, though.

BTW. My long term (ultimate) goal is to get rid of the inline styles we use a lot currently, because they cannot be overridden.

[danyeaw]

Thanks @amolenaar!

[amolenaar]

It needs a little work to fix the linting errors introduced by this PR. Linting was less strict before #2979 was merged.