Step toward #29573. Adds an independent parameter allowing the user to configure a float16 drawing buffer on a WebGPU canvas. For WebGL backends this does nothing, but my understanding is that WebGL will eventually support higher bit depth output as well.

Example:

// default to device preference
const renderer = new WebGPURenderer();

// uint8
const renderer = new WebGPURenderer( { outputType: THREE.UnsignedByteType } );

// float16
const renderer = new WebGPURenderer( { outputType: THREE.HalfFloatType } );

A quick way to test the result is to modify the TSL editor example, widen the preview renderer to at least 800px x 200px, and draw a dark gradient:

const { uniform, vec3, vec4, uv } = await import( 'three/tsl' );

const l = uv().mul( 0.01 ).r;

output = vec4( l, l, l, 1.0 );

On a good monitor (for some unspecified definition of "good"; a modern Macbook Pro display will certainly do) you will see banding with outputType: UnsignedByteType, and the banding disappears with outputType: HalfFloatType.

EDIT: I removed the screenshot, as it's difficult to get a reliable capture of the difference. I thought I had one taken correctly, but the bands seem to have disappeared (under compression on upload to Github?).