[css-view-transitions-2] Clip captured snapshots to viewport for elements

Currently when an element is snapshotted for a ViewTransition its captured in its entirety (modulo implementation constraints), see here.

But for the root the snapshot is clipped to viewport, see here.

The reason for this divergence is:

• The root is likely to be large in a lot of cases and rendering it completely is wasteful of memory/compute.
• A partially onscreen element may be brought more onscreen during the transition animations so all its content is conservatively painted.

Its unclear if the tradeoff above makes sense in all cases, maybe the author should decide this based on the transition animations they will be creating? @jakearchibald has brought up cases where the root is slightly scrolled during the transition (from a header which animates away) and needs a small margin of offscreen content to be painted.

This intersects with https://github.com/w3c/csswg-drafts/issues/8282 but I'm filing it separately so the use-case is not lost. Since https://github.com/w3c/csswg-drafts/issues/8282 could be limited to which elements participate in a transition while this is about how much content of an element is painted.

Start with the CSS View Transitions capture-the-image algorithm linked in the issue, then compare the root snapshot behavior with element snapshots. Read the related issue 8282 for overlapping scope and use the described scrolling and partially onscreen-element cases to define the desired behavior. Done requires an agreed specification change, including the clipping or painting rules.