feat(terrain): add fixed grid terrain tesselator

[charlieforward9]

Goals

• add a FAST fixed-grid terrain mesh option to @loaders.gl/terrain
• give deck.gl TerrainLayer a loader path that emits stable longitude/latitude/elevation meshes
• keep existing Martini and Delatin behavior unchanged

Changes

• add terrain.tesselator = 'grid' with configurable terrain.gridSize
• add buildGridMeshAttributes and makeGridTerrainMesh
• generate fixed-resolution indexed triangle grids with bilinear height sampling
• sample rows uniformly in Mercator-y to avoid high-latitude warping from latitude-uniform sampling
• support optional skirts on edge vertices without skewing the terrain bounding box
• export the new grid terrain helpers and types from @loaders.gl/terrain
• wire the grid path into TerrainLoader and makeTerrainMeshFromImage
• add shared terrain tests covering grid mesh size, lng/lat output, Mercator-y midpoint sampling, and skirt behavior
• add docs-ready API examples for both TerrainLoader grid usage and direct makeGridTerrainMesh() usage

Validation

• yarn lint fix
• yarn build
• yarn test-node modules/terrain/test/terrain-loader.spec.js
• yarn test-headless modules/terrain/test/terrain-loader.spec.js

Notes

• this PR intentionally does not add a standalone website/gallery demo
• the fixed grid path is aimed at FAST predictable tile density and longitude/latitude output positions for downstream renderers such as deck.gl TerrainLayer

[ibgreen-openai]

Is he goal mainly a faster tesselator?
Can we add some benchmarks (yarn test bench, yarn test bench-headless) to verify that it is worth the additional code?

[charlieforward9]

@ibgreen-openai It's not only speed, there is a warping problem in the texture that this seems to fix as well.

See visgl/deck.gl#10250 (comment)

Adding benchmarks now

[charlieforward9]

@ibgreen-openai i think its fast

[charlieforward9]

I am exploring the martini tesselator after your feedback in more detail since it does a MUCH better job of smoothing the terrain mesh out. I will continue to explore this over the coming days and decide whether this addition code path is worth it for the speed.

[ibgreen-openai]

Good idea.
I was thinking it could be good to add the size of the meshes generated to the table as well. To get a good approach with your fast mode you probably need a much finer grid which will affect memory use / rendering speed etc.
Maybe ask Codex to do an optimization pass on the existing tesselators, see if anything can be gained?

[charlieforward9]

there is a warping problem in the texture that this seems to fix as well

visgl/deck.gl#10293 fixes this

visgl/deck.gl#10294 Follows up on experiments to improve TerrainLayer loading

Closing one as I take a more responsive LoD refinementStrategy and globe cache approach for improving GlobeView initialization and flyToTransition UX