MVT: Add support for annotations

[gkjohnson]

• Create a dedicate MVT Annotations plugin that takes an overlay instance (usable via ImageOverlayPlugin)
• Track annotations (road names, locations) with a "locking" mechanism that prioritizes data from the higher LoD annotations.
• Each annotation will have a unique ID constructed for the content + quantized location
• The plugin will use tile visibility to determine which tile content is visible and track the potential annotations across the scene.
• Users can provide a "sort" callback that can be used to sort visual priority of annotations (default set to vertical positioning)
• Road names candidate locations are determined ahead of time.
• Screen space occupation grid (granularity TBD, possibly determined by annotation size) is used to detect and prevent "clashing" of annotations.
• Positioning / settling of annotations is done with raycasting spread over multiple frames.
• Text is done with a text glyph atlas and reference counting to determine whether the space is free. To start we can use a canvas set to the target resolution (no SDF) and instancing via custom shader.

[gkjohnson]

Upcoming

Possible Improvements

• WebGPU for obscurance
• Further amortized occupancy grid sorting
• Reduced annotations
• Knob for road annotation spacing?
• Tiles traversal time slicing

Load Performance

• Use a service worker for caching content
• Allow for parallelized, queue'd downloads (#1463)

Runtime Performance

• Further raycast improvements
  • Generate BVH in cartographic space, allows for quick top-down ray samples (~2x perf improvement from quick testing, but increased complexity, memory)
  • Only update sort when the camera or items have changed sufficiently - at most every few frames
  • Add option for skipping raycast, settle on ellipsoid surface

Ergonomics

• Add method for tracking download progress (possibly needed for multiple plugins with download requirements?)

Content

• There's a limit to the loadable tile content since tiles are not "split" by default on zoom in without ImageOverlayPlugin
  • This is related to #1636

Features

• Add PoI icon text "banners"
• Add more continuous motion via something like a smooth curve (remove rotation snapping around corners)
• Shift text / icons above terrain to enable more clear 3d placement (shift "up" in world space by half a radius or in relative screen direction)
• SDF font, colors for varied weights, outlines, and colors dependent on character properties
• Use a has grid to store references to the exact shapes (circles) rather than binary occupancy to make the collision checking more precise and fluid

[gkjohnson]

Plans for road annotaitons:

On Parse

• From studying maplibre: line "fragments" need to be merged into coherent annotations.
• These paths will be subsampled to a granularity based on the MVT LoD
• A unique id is determined for the road.

Anchors

• When line annotations are available a set of anchors is determined and added to the occupancy grid.
• When a new tile is loaded we associate the new road to any existing anchor that lies within the newly loaded tile. Each anchor tracks all relevant lines from all LoDs, laying out based on the highest active LoD. When an tile is removed then that line annotation is removed from the anchor - note that redundant or close anchors are NOT added.

Settling / Layout

• Just like points, all active roads are progressively settled when underlying geometry changes.
• These paths are projected to screen space an walked to lay out text on the screen, marking the occupancy grid as it goes. Sharp corners in the projected line, etc invalidate the path, causing it to hide.
• Text and glyph sizing is handled by a "driver" provided by the user.

Rendering

• Much like points, an event is fired with the final set of road annotations to be displayed in space.