StructTypeNode: Fix memory length calculations (#32375)

[holtsetio]

Fixed #32375.

Description

This change fixes the memory length calculation for the StructTypeNode so three.js allocates enough memory for custom structs.
I have also renamed the getByteBoundaryFromType() function to getAlignmentFromType() and fixed the values for mat3/mat4 according to https://webgpufundamentals.org/webgpu/lessons/webgpu-memory-layout.html

[github-actions]

📦 Bundle size

Full ESM build, minified and gzipped.

	Before	After	Diff
WebGL	350.29 83.05	350.29 83.05	+0 B +0 B
WebGPU	612.89 170.29	612.89 170.27	+1 B -17 B
WebGPU Nodes	611.49 170.01	611.49 169.99	+1 B -13 B

🌳 Bundle size after tree-shaking

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

	Before	After	Diff
WebGL	482.23 117.82	482.23 117.82	+0 B +0 B
WebGPU	683.78 186	683.78 186	+0 B +0 B
WebGPU Nodes	633.61 173.18	633.61 173.18	+0 B +0 B

[holtsetio]

Edit: Disregard this, adressed with the next commit!

To note here:
getLength() now returns the length that a buffer with a single object of this struct would occupy in memory.
In some edge cases (examples below), the returned length is too long for buffers with multiple objects of this struct. However, mitigating this is hard with the current code structure and allocating a few bytes too much is definitely better than allocating too little, as it was before..

Examples:

const testStruct = struct({
	a: { type: "vec2" },
	aa: { type: "vec2" },
	aaa: { type: "vec2" },
});

getLength() for this struct now returns 8 (32 bytes), which is correct for a buffer with a single element, e.g. instancedArray(1, testStruct);
However, for instancedArray(2, testStruct); this would result in a length of 2*32=64 bytes, even though this buffer only occupies 48 bytes.

const testStruct = struct({
	a: { type: "float" },
});

getLength() for this struct now returns 4 (16 bytes), which is correct for a buffer with a single element, e.g. instancedArray(1, testStruct);
However, for instancedArray(4, testStruct); this would result in a length of 4*32=128 bytes, even though this buffer only occupies 16 bytes.

Maybe to fully address this, there would need to be a getLength() function that also takes the number of elements of this struct into account?

[holtsetio]

Ok, I added another commit that adresses above concerns. getLength() should now also return correct lengths for structs with different alignments. getLength() now doesn't necessary return multiples of 4 anymore, but multiples of the biggest alignment value within the struct. (following https://webgpufundamentals.org/webgpu/lessons/webgpu-memory-layout.html#a-struct-array-size-alignment)

[sunag]

I think the next step is to align UniformsGroup with these changes.