feat(core): GlobeView pointer-anchored zoom (wheel + transitions)

[charlieforward9]

Summary

Adds an opt-in zoomAround: 'pointer' controller option that keeps the geographic point under the cursor/touch fixed during zoom — for wheel/pinch and for the dblclick zoom transition.

• controller.ts: new zoomAround?: 'center' | 'pointer' option on ControllerOptions.
• globe-controller.ts: GlobeState gains zoomStart / zoomEnd and a refactored zoom that uses GlobeViewport.panByGlobeAnchor when zoomAround === 'pointer'. Off-globe taps fall through to a plain zoom.
• globe-viewport.ts:
  • Shared private _getRayToGlobe helper builds the ray/sphere math used by unproject, isPointOnGlobe, and panByGlobeAnchor. JSDoc on the helper explains the contract.
  • New public isPointOnGlobe(xy, { maxDistanceRatio }).
  • New public panByGlobeAnchor(anchorLngLat, pixel) with documented edge-damping (anchor starts loosening at 0.75 × limb radius and never goes below 0.35 × strength) so near-edge anchors don't snap the camera.
• linear-interpolator.ts: replaces the prior log.warn('around not supported in GlobeView') no-op with real spherical anchoring — initializeProps stashes aroundLngLat, interpolateProps runs panByGlobeAnchor each frame. Off-globe anchors fall through to a plain LERP. This is what makes the dblclick zoom transition land on the tap point instead of the camera center.

Why

Pointer-anchored zoom matches Google Earth-style interaction. Keeping the option opt-in (default remains 'center') avoids changing existing GlobeView behavior.

Tests

• GlobeController supports pointer anchored zoom option — wheel on a globe with {zoomAround: 'pointer'} shifts longitude; default does not.
• GlobeViewport#isPointOnGlobe — center hits the globe, corner misses.
• LinearInterpolator anchors transitions on GlobeViewport — initializeProps records aroundLngLat, mid-transition longitude shifts.
• LinearInterpolator falls back to a plain LERP when the GlobeView anchor is off-globe — corner anchor leaves aroundLngLat undefined.

CI: vitest run --project browser test/modules/core/controllers/controllers.spec.ts test/modules/core/viewports/globe-viewport.spec.ts test/modules/core/transitions/linear-interpolator.spec.ts.

[coveralls]

coverage: 83.381% (-0.05%) from 83.434% — NEW-HEAT:codex/globe-anchored-zoom into visgl:master

[charlieforward9]

Added a follow-up commit (e60090a75) that extends this PR with the GlobeView transition-anchor path.

Background

The existing LinearInterpolator checks for around + makeViewport, but its globe branch is a no-op (log.warn('around not supported in GlobeView')). That meant _onDoubleClick on a globe — which already passes _getTransitionProps({around: pos}) — degraded to a plain LERP between start and end view states. Visually the camera read as zooming toward center even though the controller's own end state (from this PR's GlobeState.zoom + panByGlobeAnchor) was correctly anchored.

Change

• initializeProps: for GlobeViewport, when the anchor screen point lands on the globe (isPointOnGlobe), unproject it to lng/lat and store as aroundLngLat. Off-globe falls through to a plain LERP (matches the previous behavior).
• interpolateProps: when aroundLngLat is set and the intermediate viewport is a GlobeViewport, call panByGlobeAnchor(aroundLngLat, lerp(start.around, end.around, t)) to keep the geographic point pinned under the anchor screen point each frame. The non-globe branch is unchanged (panByPosition).
• Removed the now-unused log import.

Tests

• New LinearInterpolator anchors transitions on GlobeViewport — verifies aroundLngLat is set, aroundPosition is not set (so we don't dual-anchor), and longitude shifts mid-transition.
• New LinearInterpolator falls back to a plain LERP when the GlobeView anchor is off-globe — verifies a corner-of-canvas anchor leaves aroundLngLat undefined.

Happy to split this into a separate PR stacked on top of #10307 if you'd rather review it independently — let me know.