WebGPURenderer: Introduce Quad cache per RendeTarget texture

[sunag]

Related issue: #32690 (comment)

Description

A QuadMesh cache tends to be more efficient since a single QuadMesh sharing different output configurations and RenderTarget needs to be recompiled or have its cache key regenerated for shader reuse, making material.needsUpdate = true not so necessary.

[github-actions]

📦 Bundle size

Full ESM build, minified and gzipped.

	Before	After	Diff
WebGL	360.82 85.68	360.82 85.68	+0 B +0 B
WebGPU	632.75 175.53	632.97 175.58	+222 B +50 B
WebGPU Nodes	630.87 175.24	631.09 175.29	+222 B +49 B

🌳 Bundle size after tree-shaking

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

	Before	After	Diff
WebGL	493.02 120.25	493.02 120.25	+0 B +0 B
WebGPU	705.18 190.51	705.4 190.55	+222 B +46 B
WebGPU Nodes	654.4 177.79	654.62 177.83	+222 B +46 B