fix(stream): resume before mark-done, safe env parsing

- loadChat now calls requestResumeForAllInProgress BEFORE the
  interrupted-generation mark-done heuristic. That heuristic flips
  done=true when getTaskIdsByChatId returns empty/null, which includes
  API failures; previously that path would prevent resume from being
  attempted even when the stream was still alive server-side. Spurious
  resume requests cost one round trip and clear the fence via empty
  reply, so calling it unconditionally first is strictly safer.

- Env-var float parsing wrapped in a _float_env helper that falls back
  to default with a warning on invalid input instead of raising at
  import time. Previous code would ValueError out of the module and
  prevent socket service startup on operator typo.

Pushed back: bot flagged asyncio.Lock as unable to be stored in a
WeakValueDictionary. Verified directly on CPython — both weakref.ref
and WeakValueDictionary insertion work fine for asyncio.Lock. No
change needed.

backend/open_webui/socket/main.py

@@ -185,12 +185,19 @@ RESUME_STREAM_TTL_REFRESH_EVERY = 64
 # Replay read timeout: looser since a resume is user-blocking anyway
 # and silent timeout here is worse than a brief extra wait. Both
 # configurable for infra where Redis isn't colocated.
-RESUME_STREAM_REDIS_TIMEOUT_SEC = float(
-    os.environ.get('RESUME_STREAM_REDIS_TIMEOUT_SEC', '0.1')
-)
-RESUME_STREAM_READ_TIMEOUT_SEC = float(
-    os.environ.get('RESUME_STREAM_READ_TIMEOUT_SEC', '1.0')
-)
+def _float_env(name: str, default: float) -> float:
+    val = os.environ.get(name)
+    if val is None or val == '':
+        return default
+    try:
+        return float(val)
+    except (TypeError, ValueError):
+        log.warning(f'Invalid {name}={val!r}; using default {default}')
+        return default
+
+
+RESUME_STREAM_REDIS_TIMEOUT_SEC = _float_env('RESUME_STREAM_REDIS_TIMEOUT_SEC', 0.1)
+RESUME_STREAM_READ_TIMEOUT_SEC = _float_env('RESUME_STREAM_READ_TIMEOUT_SEC', 1.0)
 
 # Module-level circuit breaker for the streaming hot path. After N
 # consecutive Redis failures/timeouts, short-circuit seq/log calls for

src/lib/components/chat/Chat.svelte

@@ -1544,6 +1544,15 @@
 					taskIds = taskRes.task_ids;
 				}
 
+				// Request resume BEFORE marking messages done. The
+				// interrupted-generation heuristic below flips `done=true`
+				// when getTaskIdsByChatId returns empty/null, which can
+				// happen on API failure too; asking for resume first lets
+				// us recover when the stream is actually still alive.
+				// Server replies empty if no log exists, so a spurious
+				// request costs one round trip.
+				requestResumeForAllInProgress();
+
 				// If no active tasks and current message is incomplete, generation was interrupted
 				const currentMessage = history.currentId ? history.messages[history.currentId] : null;
 				if (
@@ -1555,13 +1564,6 @@
 					currentMessage.done = true;
 				}
 
-				// Resume any in-flight streams. Not gated on taskIds —
-				// that call can fail or race and is not authoritative;
-				// requestResumeForAllInProgress already filters to
-				// unfinished assistants and the server no-ops when no
-				// log exists.
-				requestResumeForAllInProgress();
-
 				await tick();
 
 				return true;