fix(stream): fence lifecycle ownership, strict done filter, narrower terminal

- onResumeStreamReplay no longer blindly clears the fence in finally.
  It stops deleting the active request_id on entry and instead checks,
  in the finally, whether it is still the active request before
  calling clearResumeFence. A newer request arriving during replay
  sets its own id; the older handler now leaves that state alone and
  lets the newer request's lifecycle drive cleanup.

- requestResumeForAllInProgress filters by `done === false` instead of
  `!done`. Legacy messages with a missing `done` field used to fan
  out as spurious resume requests on every reconnect/load; now only
  explicitly in-progress assistants get targeted.

- Terminal-event detection for TTL shortening no longer treats any
  event with `data.error` present as terminal. Restricted to
  `type === 'chat:completion' AND data.error`, matching the exact
  shape middleware.py emits when the provider actually errors out.
  A transient warning on some other event type won't shorten the log
  TTL prematurely anymore.

backend/open_webui/socket/main.py

@@ -1112,10 +1112,17 @@ async def get_event_emitter(request_info, update_db=True):
         # will no longer replay against.
         outer_type = event_data.get('type') if isinstance(event_data, dict) else None
         inner = event_data.get('data') if isinstance(event_data, dict) else None
+        # Only the specific chat:completion-with-error shape is terminal;
+        # a generic `error` field on some other event type could be a
+        # transient warning and shouldn't age the keys out early.
         is_terminal = (
             (isinstance(inner, dict) and inner.get('done') is True)
             or outer_type == 'chat:tasks:cancel'
-            or (isinstance(inner, dict) and inner.get('error'))
+            or (
+                outer_type == 'chat:completion'
+                and isinstance(inner, dict)
+                and inner.get('error')
+            )
         )
         if is_terminal and REDIS is not None and user_id and message_id:
             try:

src/lib/components/chat/Chat.svelte

@@ -724,7 +724,11 @@
 		if (expected && payload?.request_id !== expected) {
 			return;
 		}
-		resumeActiveRequestIdByMessageId.delete(messageId);
+		// Don't delete the active request_id here: a NEWER request could
+		// start during replay and set its own id. The finally below only
+		// clears the fence if we're still the active request; if a newer
+		// request has taken over, let its own lifecycle handle cleanup.
+		const ownRequestId = payload?.request_id;
 		const envelopes = Array.isArray(payload?.envelopes) ? payload.envelopes : [];
 		try {
 			for (const envelope of envelopes) {
@@ -733,8 +737,10 @@
 				await chatEventHandler(envelope);
 			}
 		} finally {
-			// Finally-clear so a mid-replay throw can't freeze live updates.
-			await clearResumeFence(messageId);
+			const current = resumeActiveRequestIdByMessageId.get(messageId);
+			if (current === ownRequestId || current == null) {
+				await clearResumeFence(messageId);
+			}
 		}
 	};
 
@@ -745,10 +751,12 @@
 	};
 
 	// Iterate all in-flight assistants so arena siblings aren't missed.
+	// Strict `done === false` so legacy messages with a missing `done`
+	// field don't get fanned out as unnecessary resume requests.
 	const requestResumeForAllInProgress = () => {
 		if (!history?.messages) return;
 		for (const message of Object.values(history.messages)) {
-			if (message && message.role === 'assistant' && !message.done) {
+			if (message && message.role === 'assistant' && message.done === false) {
 				requestResumeForMessage(message);
 			}
 		}