Fix startup connection and chunk recovery

packages/ui/src/components/chat/ChatMessage.tsx

@@ -26,13 +26,14 @@ import { filterVisibleParts } from './message/partUtils';
 import { normalizeUserDisplayParts } from './message/normalizeUserDisplayParts';
 import { flattenAssistantTextParts } from '@/lib/messages/messageText';
 import { isLikelyProviderAuthFailure, PROVIDER_AUTH_FAILURE_MESSAGE } from '@/lib/messages/providerAuthError';
+import { lazyWithChunkRecovery } from '@/lib/chunkLoadRecovery';
 import type { TurnGroupingContext } from './lib/turns/types';
 import { copyTextToClipboard } from '@/lib/clipboard';
 import { FadeInOnReveal } from './message/FadeInOnReveal';
 import { streamPerfCount } from '@/stores/utils/streamDebug';
 import { areOptionalRenderRelevantMessagesEqual, areRenderRelevantMessagesEqual, areRelevantTurnGroupingContextsEqual } from './message/renderCompare';
 
-const ToolOutputDialog = React.lazy(() => import('./message/ToolOutputDialog'));
+const ToolOutputDialog = lazyWithChunkRecovery(() => import('./message/ToolOutputDialog'));
 
 const EXPANDED_TOOLS_CACHE_MAX = 4000;
 const expandedToolsStateCache = new Map<string, Set<string>>();

packages/ui/src/components/chat/MarkdownRenderer.tsx

@@ -1,4 +1,5 @@
 import React from 'react';
+import { lazyWithChunkRecovery } from '@/lib/chunkLoadRecovery';
 
 // Thin lazy wrapper around the heavy MarkdownRenderer implementation.
 // The full implementation (marked, react-markdown, beautiful-mermaid,
@@ -8,11 +9,11 @@ import React from 'react';
 export type { MarkdownVariant } from './MarkdownRendererImpl';
 
 
-const MarkdownRendererLazy = React.lazy(() =>
+const MarkdownRendererLazy = lazyWithChunkRecovery(() =>
   import('./MarkdownRendererImpl').then((m) => ({ default: m.MarkdownRenderer }))
 );
 
-const SimpleMarkdownRendererLazy = React.lazy(() =>
+const SimpleMarkdownRendererLazy = lazyWithChunkRecovery(() =>
   import('./MarkdownRendererImpl').then((m) => ({ default: m.SimpleMarkdownRenderer }))
 );