Cleanup flocking example to use composition

[tychedelia]

Flocking is now built from the generic ops, so the built-in flock goes.

• Removed: p.flock(), particles_flock, particles_kernel_flock, the flock.wgsl kernel, and its FFI constructor.
• flocking_gpu.py is now a recipe with no shader code: find neighbors, sum what each rule steers toward, apply steering with a mask for boids that have no neighbors, then add the force to velocity and limit the speed.

It's slower than the fused kernel: about 3.5 ms vs 1.6 ms at 20k boids, and 8 ms vs 4 ms at 100k. Most of that is the per-dispatch cost of ~30 passes, not GPU work, so batching submits is the next thing to look at. Anyone who needs the speed can still write a small kernel that reads neighbors directly.

[github-actions]

Visual regression: 1 changed of 43 cases

case	status	detail
py_flocking_gpu	❌ changed	28.6% of pixels differ

If these changes are intentional, add the deliberate-rendering-change label.

_{baseline 496b5c8bd8 · adapter llvmpipe (LLVM 20.1.2, 256 bits) (Vulkan, Mesa 25.2.8-0ubuntu0.24.04.4 (LLVM 20.1.2))}

Open the visual report · workflow run

[catilac]

this is really really cool, @tychedelia. I will play with this API after we get this all merged. This looks good though. It looks like flocking_gpu_flock.wesl and flocking_gpu_integrate.wesl can be deleted now?