Fix gesture hit-testing for transformed controls in virtualized/cached scrolls

[LennoxP90]

PR: Fix gesture hit-testing for transformed controls in virtualized/cached scrolls

Fixes #302.

Problem

A control with a transform (ScaleX/ScaleY, Rotation*, skew, flip) renders correctly but pointer
events never reach its children — the transform is applied to rendering but ignored by gesture
hit-testing. This breaks the standard inverted-list pattern (a ScaleY="-1" SkiaScroll with
counter-flipped cells for a newest-at-bottom chat list) inside virtualization: cells render upright,
but taps on links/buttons inside a cell do nothing. See the linked issue for the full root-cause.

Fix

Gesture hit-testing only — no rendering changes:

1. SkiaControl.EnsureRenderTransformMatrixForGestures() (new): recomputes RenderTransformMatrix
   from the current DrawingRect for transformed controls. Called before the inverse-map in
   ProcessGestures and in IsGestureForChild. Fixes the stale/Identity matrix on node-attached/cached
   controls and the stale pivot on virtualized controls (whose DrawingRect moves as they scroll).
2. SkiaScroll.ProcessGesturesForPlane: map the gesture point and the forwarded MappedLocation
   by the scroll's transform (the virtualized planes path overrides ProcessGestures and never reaches
   the base entry inverse-map).
3. SkiaControl.ProcessGestures (registered-listener branch): hit-test transformed controls with
   the entry-mapped MappedLocation (consistent with the rendering-tree branch); the non-transformed
   path is kept exactly as before.
4. SkiaLabel.ProcessGestures: span (link) hit-testing uses the entry-mapped MappedLocation
   instead of the raw args.Event.Location.

Three files: SkiaControl.Shared.cs, SkiaScroll.Planes.cs, SkiaLabel.cs.

No regression for non-transformed UI

Changes 1–3 are gated on HasTransform; change 4 is identical to the previous behaviour when nothing
in the ancestry is transformed, because MappedLocation == args.Event.Location in that case (the root
sets MappedLocation = touchLocation and the entry inverse-map is a no-op without a transform). So a
normal, non-transformed list takes byte-identical code paths.

Testing

• Android (net10): verified that tapping a URL link inside a cell of an inverted
  (ScaleY="-1") virtualized SkiaScroll now fires (opens the browser); without the change the tap is
  detected on the cell but never reaches the label's link span.
• Non-transformed regression check: the same list with the transforms removed still routes taps
  correctly (the gated paths fall through to the original behaviour).
• The change is platform-agnostic C#; I could not run the iOS/MacCatalyst matrix on a Windows-only host.

Notes / for discussion

• EnsureRenderTransformMatrixForGestures() recomputes the matrix per gesture event for transformed
  controls (a single cheap matrix build). If preferred, this could be made dirty-tracked / cached and
  invalidated on transform or layout change — happy to adjust.
• The registered-listener branch was previously inconsistent with the rendering-tree branch (raw vs
  mapped coordinates); this aligns them for transformed controls while leaving the common path intact.