Multi-turn conversation, stop generation, SSE fix, and UI improvements

- Multi-turn context: all council stages now receive conversation history
  (user messages + Stage 3 chairman responses) for coherent follow-ups
- Stop generation: abort streaming mid-request, recover query to input box
- SSE parsing: buffer-based chunking to prevent JSON split across packets
- Atomic storage: user + assistant messages saved together after completion,
  preventing dangling messages on abort
- GFM markdown: tables, strikethrough via remark-gfm plugin + table styles
- Performance: memo user messages and completed assistant messages, only
  re-render the active streaming message
- Model config: gpt-5.2, claude-opus-4.6 as chairman
- Always show input box for multi-turn conversations

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

1 files changed, 2 insertions, 1 deletions

diff --git a/frontend/src/components/Stage3.jsx b/frontend/src/components/Stage3.jsx
index 9a9dbf7..f63a442 100644
--- a/frontend/src/components/Stage3.jsx
+++ b/frontend/src/components/Stage3.jsx
@@ -1,4 +1,5 @@
 import ReactMarkdown from 'react-markdown';
+import remarkGfm from 'remark-gfm';
 import './Stage3.css';
 
 export default function Stage3({ finalResponse }) {
@@ -14,7 +15,7 @@ export default function Stage3({ finalResponse }) {
           Chairman: {finalResponse.model.split('/')[1] || finalResponse.model}
         </div>
         <div className="final-text markdown-content">
-          <ReactMarkdown>{finalResponse.response}</ReactMarkdown>
+          <ReactMarkdown remarkPlugins={[remarkGfm]}>{finalResponse.response}</ReactMarkdown>
         </div>
       </div>
     </div>