Fix dropped full_conversation replies: don't early-return on rawText

The extension's full_conversation reply always carries `rawText` in its
result. handleExtensionMessage returned early when RawSSEOutputPath was
unset, so the reply never reached pending[msg.ID] and every browser-mode
chat timed out (240s) with an empty response. Now the raw-SSE dump is
purely optional and dispatch always continues to the pending channel.

Also sync the validated gpt-extension/background.js into the Docker build
source so HF builds the working extension.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

chatgpt-free-api/extension.go

@@ -88,12 +88,14 @@ func handleExtensionMessage(msg WSMessage) {
 	if msg.Result != nil {
 		if m, ok := msg.Result.(map[string]any); ok {
 			if raw, ok := m["rawText"].(string); ok {
-				if cfg.RawSSEOutputPath == "" {
-					return
-				}
-				err := os.WriteFile(cfg.RawSSEOutputPath, []byte(raw), 0644)
-				if err != nil {
-					log.Printf("ERROR WRITING RAW SSE: %v", err)
+				// Optionally dump the raw SSE for debugging. This must NOT
+				// short-circuit dispatch: the same reply still has to reach the
+				// pending[msg.ID] channel below, otherwise full_conversation
+				// responses are silently dropped and the caller times out.
+				if cfg.RawSSEOutputPath != "" {
+					if err := os.WriteFile(cfg.RawSSEOutputPath, []byte(raw), 0644); err != nil {
+						log.Printf("ERROR WRITING RAW SSE: %v", err)
+					}
 				}
 			}
 		}

chatgpt-free-api/gpt-extension/background.js

@@ -900,7 +900,31 @@ async function handleFullConversation(msg) {
       return;
     }
 
-    const text = await convResp.text();
+    // Read the SSE stream incrementally instead of await convResp.text().
+    // ChatGPT keeps the HTTP connection alive after emitting `data: [DONE]`, so
+    // .text() (which resolves only on connection close) hangs forever. We stop as
+    // soon as we see the stream-completion marker, with a hard safety deadline.
+    let text = '';
+    {
+      const STREAM_DEADLINE_MS = 180000; // overall cap so a stalled stream can't hang the SW
+      const deadline = Date.now() + STREAM_DEADLINE_MS;
+      try {
+        const reader = convResp.body.getReader();
+        const decoder = new TextDecoder();
+        while (true) {
+          const { done, value } = await reader.read();
+          if (done) break;
+          text += decoder.decode(value, { stream: true });
+          // Stop once the server signals completion (it won't close the socket).
+          if (text.includes('[DONE]')) { try { await reader.cancel(); } catch (e) {} break; }
+          if (Date.now() > deadline) { try { await reader.cancel(); } catch (e) {} break; }
+        }
+        text += decoder.decode();
+      } catch (e) {
+        // Fall back to whatever we managed to read; parseSSEFinal handles partials.
+        console.error('[ChatGPT Bridge] stream read error:', e);
+      }
+    }
     let convId = '';
     const parseSSEFinal = (raw) => {
       let snapshot = '';