fix: defer LLM stream connect timeout to after HTTP request is sent

packages/opencode/src/session/llm.ts

@@ -494,13 +494,18 @@ const live: Layer.Layer<
                       })
                   }, currentTimeoutMs())
                 }
-                arm()
                 return {
                   ctrl,
                   timeoutFailure,
                   timeoutError() {
                     return timeoutError
                   },
+                  // Start the connect timeout. Called after run() returns so the
+                  // timer only measures actual network/provider wait time, not
+                  // the internal setup work (provider lookup, config, plugins).
+                  startTimeout() {
+                    arm()
+                  },
                   resetTimeout(event: Event) {
                     if (!providerProgressed && !isProviderProgressEvent(event)) return
                     providerProgressed = true
@@ -519,6 +524,12 @@ const live: Layer.Layer<
 
             const result = yield* run({ ...input, abort: request.ctrl.signal })
 
+            // Arm the connect timeout now that the HTTP request has been sent.
+            // Previously this was called during request object creation, which
+            // meant setup time (provider lookup, config, plugin hooks) ate into
+            // the 30s window, causing premature timeouts.
+            request.startTimeout()
+
             // This is a silent-stream timeout: it limits how long we wait for
             // the next provider event, not the total model runtime.
             return Stream.fromAsyncIterable(failOnTimeout(result.fullStream, request), (e) =>