refactor: streamline handling of <details> tags in markdown parsing

- Updated the assistant message widget to rely on the markdown parser for <details> tags, eliminating manual tag management for tool_calls and reasoning types.
- Enhanced the details builder to ensure proper handling of <details> elements during streaming, preventing character flashing.
- Cleaned up the code by removing redundant comments and consolidating logic for improved readability and maintainability.

lib/features/chat/widgets/assistant_message_widget.dart

@@ -720,26 +720,14 @@ class _AssistantMessageWidgetState extends ConsumerState<AssistantMessageWidget>
       return const SizedBox.shrink();
     }
 
-    // Always hide tool_calls blocks; tiles render them separately.
+    // Note: The markdown parser now handles <details> tags (including type="reasoning"
-    String cleaned = content.replaceAll(
+    // and type="tool_calls") via a custom block syntax, so they won't be rendered as
-      RegExp(
+    // plain text during streaming. This prevents character flashing.
-        r'<details\s+type="tool_calls"[^>]*>[\s\S]*?<\/details>',
+
-        multiLine: true,
+    // We still clean raw reasoning tags (<think>, <reasoning>) as a fallback.
-        dotAll: true,
+    // The server normally converts these to <details> format, but raw mode or
-      ),
+    // direct API responses might still use them.
-      '',
+    String cleaned = content
-    );
-    // Also hide reasoning details blocks if any slipped into text
-    cleaned = cleaned.replaceAll(
-      RegExp(
-        r'<details\s+type="reasoning"[^>]*>[\s\S]*?<\/details>',
-        multiLine: true,
-        dotAll: true,
-      ),
-      '',
-    );
-    // Remove raw <think>...</think> or <reasoning>...</reasoning> tags in text
-    cleaned = cleaned
         .replaceAll(
           RegExp(r'<think>[\s\S]*?<\/think>', multiLine: true, dotAll: true),
           '',
@@ -753,10 +741,6 @@ class _AssistantMessageWidgetState extends ConsumerState<AssistantMessageWidget>
           '',
         );
 
-    // Note: The markdown parser now handles <details> tags via a custom block syntax,
-    // so they won't be rendered as plain text during streaming. This prevents the
-    // character flashing issue.
-
     // Process images in the remaining text
     final processedContent = _processContentForImages(cleaned);
 

lib/shared/widgets/markdown/markdown_config.dart

@@ -763,9 +763,11 @@ class _DetailsBuilder extends MarkdownElementBuilder {
 
   @override
   Widget? visitElementAfter(md.Element element, TextStyle? preferredStyle) {
-    // The details element should not be rendered as markdown during streaming.
+    // Details elements with type="reasoning" or type="tool_calls" should not be
-    // Instead, it's handled by the ReasoningParser in assistant_message_widget.
+    // rendered as markdown during streaming. They are handled by:
-    // Return empty widget to prevent flashing.
+    // - ReasoningParser for reasoning blocks (creates thinking tiles)
+    // - ToolCallsParser for tool_calls blocks (creates tool execution tiles)
+    // Return empty widget to prevent character flashing during streaming.
     return const SizedBox.shrink();
   }
 }