WebGLBackend: Optimize vertex state definition.

[Mugen87]

Related issue: #30560

Description

While investigating #30560, I have noticed a lot of redundant vertex state definitions in the WebGL backend of WebGPURenderer.

VAOs and index buffers can be cached and bindVertexArray() and bindBuffer() must only be executed if there is a state change.

[github-actions]

📦 Bundle size

Full ESM build, minified and gzipped.

	Before	After	Diff
WebGL	336.51 78.38	336.51 78.38	+0 B +0 B
WebGPU	521.78 144.83	522.05 144.92	+273 B +90 B
WebGPU Nodes	521.25 144.72	521.52 144.81	+273 B +90 B

🌳 Bundle size after tree-shaking

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

	Before	After	Diff
WebGL	465.51 112.23	465.51 112.23	+0 B +0 B
WebGPU	592.15 160.44	592.43 160.53	+273 B +92 B
WebGPU Nodes	547.28 149.88	547.56 149.98	+273 B +93 B

[Mugen87]

These are screenshots from SpectorJS that show the WebGL state changes in the live example from #30560. Notice how the redundant bindVertexArray() commands are gone in the second screenshot.

WebGPURenderer with WebGL2:

With this PR:

[Mugen87]

FYI: The index is no part of the VAO anymore similar to WebGLRenderer. If not doing this, there are scenarios where the index buffer is not correctly bound which results in a "no elements buffer bound" warning in WebGL. If that happens, the affected primitives won't render.