WebGLRenderer: Optimize VSM shadow code

[mrdoob]

Description

Changes

• Better variable names (mean, variance, d, p_max) for clarity
• Removed redundant outer clamp() - result is already guaranteed to be in [0,1]
• Pre-computed constant (0.65 instead of 0.95 - 0.3)
• Early return optimization for fully lit pixels
• Fixed variance epsilon from 0.00000 to 0.0000001 to actually prevent division by zero

Benefits

• Slightly better performance: Fewer operations per pixel
• Improved numerical stability: Proper epsilon prevents division by zero and NaN issues
• More maintainable: Clearer variable names and better code structure
• Identical visual output: All changes are mathematically equivalent to the original

[github-actions]

📦 Bundle size

Full ESM build, minified and gzipped.

	Before	After	Diff
WebGL	355.14 86.33	355.01 86.3	-136 B -26 B
WebGPU	609.62 170.99	609.62 170.99	+0 B +0 B
WebGPU Nodes	608.23 170.72	608.23 170.72	+0 B +0 B

🌳 Bundle size after tree-shaking

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

	Before	After	Diff
WebGL	486.77 121.07	486.63 121.05	-136 B -26 B
WebGPU	678.47 186.28	678.47 186.28	+0 B +0 B
WebGPU Nodes	620.47 169.54	620.47 169.54	+0 B +0 B

[mrdoob]

Maybe #32180 makes VSM shadows redundant too...

[Mugen87]

VSM can blur the shadow map as a whole which a great quality advantage over PCF. Also in #32180, you see the noise structure especially with larger radii at the shadow edges because there is no denoise or temporal filtering in place. Depending on the scene and how closely you look at the shadows, that could be considered as a quality issue (and a reason for using VSM).