Selection outline should respect transform

[JonnyBurger]

In the centered-solid composition, we have a <Solid> with a rotation of 45 degrees.

but the outline shows no rotation.

We should think about how we can support showing an outline on transformed elements.

In web-renderer, we have a sophisticated algorithm for recursively finding out the transform of something, but it actually leads to layout recalculation. This is undesired.

This might eventually be too complicated to realize. I don't think we can find the 3D transform that is applied to an element, composed of it's own and it's parent.

So we should also ask ourselves - where to draw the line?

[JonnyBurger]

Plan proposal:

1. Fork jogibear9988/getBoxQuadsPolyfill and use it as a reference implementation, but do not vendor it as a global polyfill.
2. Extract the useful geometry logic into a Remotion-owned ponyfill API, for example getBoxQuadsPonyfill(element, options), so we never patch Node.prototype or override browser globals.
3. Keep the initial Studio usage narrow: measure only the currently selected outline target, preferably via the existing refForOutline, not every sequence in the timeline.
4. Use the ponyfill to return a 2D DOMQuad for the selected element relative to the preview/portal container, then render the selection outline as an overlay polygon/quad.
5. Treat native element.getBoxQuads?.() as an optional fast path where available, but normalize all callers through our ponyfill-shaped function.
6. Avoid web-renderer-style DOM mutation. The approach may do layout reads (getBoundingClientRect(), offsetWidth, computed style, etc.), but it should not neutralize transforms by writing styles before measuring.
7. Batch reads before writes: compute the quad first, then update overlay state. Recompute only on selection changes, frame changes, zoom/pan/resize, and relevant hot reload updates.
8. Start with HTML elements and common 2D transforms. Keep SVG/text/multi-fragment/3D/perspective support as best-effort or follow-up scope unless the forked algorithm gives us those safely.
9. Fallback behavior: if quad calculation fails or hits an unsupported case, keep the current axis-aligned outline rather than breaking selection UI.

This gives us transformed outlines without relying on global monkey-patching, while keeping the performance risk bounded to one selected element.