BatchedMesh: update example, fix depth conversion & auxiliary buffer

[sciecode]

Related issue: #27202, #27213

SortUtils: minor changes - smaller memory footprint.
webgl_mesh_batch: converts depth to uint32 by bit-cast instead of normalization.
webgl_mesh_batch: correctly maintains auxiliary buffer between calls, based on geometry count.

Simply normalizing the float depth is not enough to correctly sort, as the fractional part gets truncated - producing incorrect ties. If float values are strictly positive, a simple bit-cast is sufficient to solve this. If negative values are possible you can use the following conversion:

const f32 = new Float32Array( 1 );
const u32 = new Uint32Array( f32.buffer );
const convert = (val) => {
   f32[0] = val;
   return u32[0] ^ ( - ( val >>> 31 ) | 0x80000000 );
}

[gkjohnson]

Thanks for the fixes! I added a couple comments.

[sciecode]

Thanks for the fixes! I added a couple comments.

Addressed both comments, let me know what you prefer. I'll make the changes.

[sciecode]

Should be good to go.
@gkjohnson

[gkjohnson]

Awesome, thanks!