TSL: Replace Proxy approach with prototypes

[sunag]

Related issue: #30849

Description

The new approach should improve performance overall, both in compilation and updates, as well as make debugging easier.

The most affected part is array access through numeric constants, which was limited between 0–32. Luckily, it’s uncommon to use constants for values larger than that. In all cases, it’s better to use a.element(b), which accepts dynamic values.

[github-actions]

📦 Bundle size

Full ESM build, minified and gzipped.

	Before	After	Diff
WebGL	338.91 79.1	338.91 79.1	+0 B +0 B
WebGPU	576.19 158.93	576.28 158.83	+87 B -104 B
WebGPU Nodes	574.8 158.69	574.89 158.59	+87 B -103 B

🌳 Bundle size after tree-shaking

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

	Before	After	Diff
WebGL	470.7 113.91	470.7 113.91	+0 B +0 B
WebGPU	646.81 174.91	646.9 174.9	+87 B -9 B
WebGPU Nodes	600.99 164.09	601.07 164.05	+87 B -43 B

[cmhhelgeson]

Why does switching from Proxy to standard objects improve performance? Are proxy gets/sets slower than regular object get/sets?

[sunag]

Why does switching from Proxy to standard objects improve performance? Are proxy gets/sets slower than regular object get/sets?

Proxy introduces an extra verification layer — imagine that for every value we want to obtain or function we want to call, it has to go through an intermediate interpreter.

We are already using a .getSelf() function to avoid Proxy in uniform update events, which should also be removed in this PR.

[mrdoob]

Getters and setters were too slow in 2013 so I ended up having to do this: 972c852
I was hoping nowadays things would have improved and we could even use Proxy now, but I guess not...

[sunag]

I’m ignoring the property shortcuts width->x, height->y, depth->w because they’re causing conflicts with some nodes and apparently aren’t used in any example.