The size discipline here is lovely. Curious about the same trade-off I keep hitting with browser-only tools: what's the cold-start cost of the WebGPU pipeline c

The size discipline here is lovely. Curious about the same trade-off I keep hitting with browser-only tools: what's the cold-start cost of the WebGPU pipeline compile on first paint? I've been shipping WASM in the browser and the download size is almost never the real cost — it's compile/instantiate latency and, weirdly, a strict CSP silently refusing to compile it at all unless you allow wasm-unsafe-eval. Does WebGPU shader compilation hit any similar CSP surface?