[eric] perf: mass-delete parallel-detaches browsers too (removeBrowserCardsCleanly); flat latency, same crash-safe ordering

frontend/src/app/pages/Dashboard/hooks/interaction/deleteSelectedCards.ts

@@ -1,7 +1,7 @@
 import { closeSession } from '@/shared/state/agentsSlice';
 import { removeNote, removeWorkflowCard, closeWorkflowsHub, recordClosedCard } from '@/shared/state/dashboardLayoutSlice';
 import { closeWorkflowCard } from '@/shared/state/workflowsSlice';
-import { removeBrowserCardCleanly } from '@/shared/browserTeardown';
+import { removeBrowserCardsCleanly } from '@/shared/browserTeardown';
 import { removeViewCardCleanly } from '@/shared/viewTeardown';
 import type { AppDispatch } from '@/shared/state/store';
 import type { CardType } from '../state/useDashboardSelection';
@@ -36,6 +36,6 @@ export function deleteSelectedCards(selectedIds: Map<string, CardType>, dispatch
   // "non-existent mailbox" errors and SIGSEGVs the GPU/browser process (the mass-delete self-quit).
   void (async () => {
     for (const id of viewIds) await removeViewCardCleanly(id, dispatch);
-    for (const id of browserIds) await removeBrowserCardCleanly(id, dispatch);
+    await removeBrowserCardsCleanly(browserIds, dispatch);
   })();
 }

frontend/src/shared/browserTeardown.ts

@@ -38,3 +38,17 @@ export async function removeBrowserCardCleanly(
   forgetBrowser(browserId);
   dispatch(removeBrowserCard(browserId));
 }
+
+// Batch remove: detach every browser's CDP in PARALLEL (order-independent, the only invariant is
+// all debuggers gone before any unmount), THEN remove them. Serial detach cost ~72ms/browser, so a
+// 20-card multi-select delete lagged over a second; this keeps the crash-safe ordering but flat.
+export async function removeBrowserCardsCleanly(
+  browserIds: string[],
+  dispatch: Dispatch,
+): Promise<void> {
+  await Promise.allSettled(browserIds.map((id) => detachBrowserCdp(id)));
+  for (const id of browserIds) {
+    forgetBrowser(id);
+    dispatch(removeBrowserCard(id));
+  }
+}