Sam Valladares ba5a28d5f9 fix(dashboard): close remaining BackdropEngine audit findings (M3 H1-H4 + mediums) ... Clears the rest of the MiniMax M3 audit backlog (all verified, svelte-check green): - H1: Canvas2D fallback now honors decay (inward pull+damp) and birth (outward bloom) events, not just firewall — full event parity with the WGSL path for pre-iOS-26 / no-WebGPU users. - H2: noise/flow time is now an `activeT` accumulator advanced by dt only while drawing, so a backgrounded tab no longer fast-forwards the curl flow and snaps it to a new direction on resume (was wall-clock `(now-startedAt)`). Both the WebGPU and Canvas2D paths share it. - H3: prefers-reduced-motion is now reactive ($state + a matchMedia change listener) so toggling the OS setting mid-session takes effect next frame (was a const snapshot at init). - H4: documented that `count` is read once at boot (buffer sizing); `intensity` is the live knob. - M3: cache the Canvas2D context once in bootFallback (was getContext per frame). - M4: coalesce resize bursts to one resize/frame via rAF (was thrashing the swapchain config + 2D backing store on every drag event). - M7: register device.lost BEFORE pipeline creation (was after the handle, leaving a race window where a loss mid-build never fired the fallback). - L2/L4/L1: drop the empty onMount return (single onDestroy teardown); reset lastT in bootFallback so a WebGPU recovery sees no giant dt; document the intentionally-non-reactive animation state. svelte-check 0 errors / 967 files. Cosmetic-only items left (M1 dead DPR slot, M2 dead vel.w, M5 sizePx naming, L3 alpha-mode smell) — no behavior, deferred. Memory Cinema / Graph3D untouched. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>		2026-06-27 15:35:05 -05:00
..
dashboard	fix(dashboard): close remaining BackdropEngine audit findings (M3 H1-H4 + mediums)	2026-06-27 15:35:05 -05:00