Components are widgets

[dhardy]

Alternative to #315: components are widgets. This resolves ambiguities over whether something is a widget or a component. It is sometimes a little inefficient (storing a WidgetId for things like a label which don't need one), but this is not significant.

• Replace Layout and Mark components with real widgets (removes kas::component module)
• Allow widgets introduced by layout syntax to be enumerated by WidgetChildren and configured
• Allow embedding of widgets within layout syntax (currently these are boxed; given something like RFC 2524 they could be placed inline, but again this inefficiency is not important)
• Remove support for glob (*) from layout syntax; this is less important (since widgets may be embedded) and potentially confusing given that layout syntax also declares child widgets

Example: the stopwatch can now be declared as follows, with the buttons inlined into the layout:

        #[widget{
            layout = row: [
                self.display,
                TextButton::new_msg("&reset", MsgReset),
                TextButton::new_msg("&start / &stop", MsgStart),
            ];
        }]
        struct {
            #[widget] display: impl HasString = Frame::new(Label::new("0.000".to_string())),
            saved: Duration = Duration::default(),
            start: Option<Instant> = None,
        }

Now, child widgets must appear as #[widget] fields in one of two cases:

• The field is directly read/written to by event handlers
• Construction of the field depends on constructor parameters or moved values

---

Inherits some other commits from #315: removing Widget: 'static bound; preferring dyn Widget over W: Widget in event-management code; some clean-up. Closes #315.

The Label widget now allows replacing the text class.

Minor change to layout of examples/data-list*

[dhardy]

This inherits default implementations:

• set_rect just assigns to self.core.rect (fine)
• find_id returns self.id() if self.core.rect contains coord — okay, but since this widget never handles click events (or any events), maybe it's better to return None — that way, "forgetting" to use #[widget] on a field to configure the widget is harmless

The same applies to Label too. Not a change for this PR, but possibly a future one.

[dhardy]

This idea only works if operations on invalid ids are changed to not panic. (The fix I just pushed is because a label was not configured and whose id is checked for disabled status when drawn, resulting in a crash due to use of an unconfigured id.)