Shared view transform — zoom, pan_x, pan_y

A camera zoom about the frame centre, then a pan. Applied by every coloured scene: fragment_field, swarm, the line systems (parametric_curve / lsystem / star_pattern, and spectrum), and reaction_diffusion and attractor.

• On the line, swarm, and attractor scenes, zoom > 1 moves the camera in (geometry bigger); zoom = 1 is no zoom. pan_* shift in world units. Try zoom = "1 + bass * 0.6" for a kick-driven pump.
• On the fragment field and reaction-diffusion, zoom scales the sample coordinates of the present pass, so a higher zoom shows more of the field (the opposite sense to the geometry scenes, kept for the shipped fragment presets); pan_x / pan_y slide the sampled window.

Reaction-diffusion’s field is finite but toroidal, and that is what makes the view transform usable on it. The simulation runs on a fixed grid whose edges wrap — growth leaving the right edge re-enters on the left — so the field is seamless even though it is not infinite, and the present pass samples it that way too:

• pan_x / pan_y are a seamless infinite scroll. Pan as far as you like in either axis; the field repeats, and there is no edge to fall off.
• zoom > 1 tiles the field rather than running out of it. At zoom = 1.4 you see the whole field plus a wrapped border on each side, joined seamlessly.

Before that change the present sampler clamped, so anything past the field edge was the boundary row smeared outward — zoom > 1 produced vertical bars and rectangular blocks, and any real pan_* walked off into them. That is why the shipped reaction_* presets are pinned near zoom = 0.99 with a whisper of pan, and why those pins are now unnecessary.