Each base change triggered an immediate clone+postMessage on top of the
already-debounced pre-bake, costing ~5-15ms per click through history.
Folding the base back into each composite call (atomic per-message) lets
the debounced pre-bake absorb the clone cost on its own schedule and
sidesteps a race where concurrent bakes with different bases could leave
a stale base in the worker-side cache.