PMREMGenerator: Use highp precision in GGX VNDF shader

[mrdoob]

Related issue: #32202

Description

Changes GGX VNDF shader precision from mediump to highp to attempt to resolve IBL darkening on Android devices (Pixel 9, Meta Quest 3).

The GGX shader performs bit manipulation (Van der Corput radical inverse) and accumulates 256+ samples. On Android GPUs, mediump provides only ~10 bits of mantissa precision, causing visible lighting errors.

Also reduces GGX_SAMPLES from 512 to 256 for better performance.

[github-actions]

📦 Bundle size

Full ESM build, minified and gzipped.

	Before	After	Diff
WebGL	356.24 86.47	356.24 86.48	-4 B +8 B
WebGPU	610.57 171.28	610.57 171.28	+0 B +0 B
WebGPU Nodes	609.18 171.02	609.18 171.02	+0 B +0 B

🌳 Bundle size after tree-shaking

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

	Before	After	Diff
WebGL	488.03 121.24	488.02 121.25	-4 B +8 B
WebGPU	679.51 186.54	679.51 186.54	+0 B +0 B
WebGPU Nodes	621.23 169.71	621.23 169.71	+0 B +0 B