Optimize Gaussian Splat loading

[bhouston]

I profiled the Gaussian Splat loaders this morning and found a few slow spots that were easy to optimize:

• I was manually clamping bytes unnecessarily when I could just use a UInt8ClampedArray.
• Inlined some math, rather than using helpers.
• Removed data access indirection.
• Inlined some small loops.
• Added lookup tables when there is 1024 or less values for the stored data in the SPZLoader.

Overall, this speeds up the Gaussian Splat loaders so that they barely show up in the profiler outside of the SPZLoader's dependency on fflate (which is outside of my optimization scope.)

[github-actions]

📦 Bundle size

Full ESM build, minified and gzipped.

	Before	After	Diff
WebGL	366.31 86.86	366.31 86.86	+0 B +0 B
WebGPU	696.95 192.4	696.95 192.4	+0 B +0 B
WebGPU Nodes	694.94 192.1	694.94 192.1	+0 B +0 B

🌳 Bundle size after tree-shaking

Minimal build including a renderer, camera, empty scene, and dependencies.

	Before	After	Diff
WebGL	509.7 123.83	510.37 124.05	+670 B +224 B
WebGPU	771.58 207.3	772.25 207.52	+670 B +224 B
WebGPU Nodes	720.54 194.63	721.21 194.86	+670 B +228 B