feat(@typegpu/noise): Stateless PRNGs

Right now, our randf.* utilities imported from @typegpu/noise are all stateful. To illustrate this, let's look at randf.sample():

const a = randf.sample();
const b = randf.sample();

a and b are different pseudo-random numbers, even though the call to randf.sample() is the same. There is hidden state, updated on each call to randf.sample(). This state is private, meaning it's shared between calls to these functions on the same thread. The issue mostly arises when using multiple random utilities, like perlin-noise, in conjunction with the more low-level randf.sample().

function main(uv: d.v2f) {
  'use gpu';
  randf.seed(time.$);

  const noise = perlin2d.sample(uv); // perlin2d.sample resets the seed
  const factor = randf.sample(); // no longer seeded based on time
}

Stateless generators

[!NOTE]
The proper API is still yet to be finalized

import { randf } from '@typegpu/noise/stateless';

function main() {
  'use gpu';
  const state = d.ref(randf.createState());
  randf.seed(state, time.$);

  const value = randf.sample(state);
  const direction = randf.randOnUnitSphere(state);
}

The generator slot would be shared between the stateful and stateless implementations.

Start by reviewing the existing randf.* utilities and perlin2d.sample() entry points in @typegpu/noise, then compare them with the proposed @typegpu/noise/stateless API. Done means the stateless generators have a finalized API, explicit state, and compatible generator-slot behavior alongside the stateful implementation.