[elements] Derive a DOM tree node's shape in one place in DOMTreeWidget

The view computed "does this node have children, is it expanded, which
children are shown, does it need a closing tag" twice, once to render
and once to count rows for maxRowsShown, and the "not collapsible" and
"being edited as HTML" predicates were spelled out in four and three
places across the view and the presenter.

- resolveNodeShape() in the view feeds both renderNode() and
  countVisibleRowsForNode(); isNotCollapsible() and
  isBeingEditedAsHTML() are module-level helpers used by the presenter
  as well.
- nodeHasVisibleChildren() / getVisibleChildren() take a TreeOptions
  object that both the view input and the presenter satisfy, instead
  of the same three or four positional arguments at every call site.
- The three opening/closing-tag handler pairs (mousemove, dragover,
  drop) become factories parameterized by isClosingTag, and the
  per-node copy of the InterceptBindingDirective is gone.
- ViewInput is module-private and the presenter always provides every
  callback, so they are now required. That removes the optional
  chaining and the dead expandRoot / InitialChildrenLimit fallbacks in
  the view. The view no longer receives the presenter itself; the
  three things it used it for are explicit inputs (onRevealInTopLayer,
  onSetMultilineEditing, visibleWidth).

No change in behavior.

Bug: 407751692
Change-Id: I249f0425320700fe751bc25d7dde03e793fbd79b
Reviewed-on: https://chromium-review.googlesource.com/c/devtools/devtools-frontend/+/8568723
Reviewed-by: Simon Zünd <szuend@chromium.org>
Auto-Submit: Alex Rudenko <alexrudenko@chromium.org>
Commit-Queue: Simon Zünd <szuend@chromium.org>