fix: reconcile live scheduled resume cancellation

src/wf_api/runs.py

@@ -221,14 +221,12 @@ class WorkflowRunApi:
                 store=store,
             )
         except asyncio.CancelledError:
-            # Shutdown drain cancelled the execution mid-flight (or the
-            # caller went away): fence, don't release. The durable
-            # executing mark and the ACTIVE attempt stay exactly as a
-            # crash mid-resume would leave them, so restart recovery
-            # fails the run closed instead of presenting a half-resumed
-            # run as safe to retry. Cancellation is never swallowed.
+            # Shutdown cancellation keeps the durable crash shape for
+            # restart recovery. A caller cancellation while the service is
+            # still live is reconciled for this run immediately, so its
+            # ambiguous failure frees capacity without becoming retryable.
             if slot_held and gate is not None:
-                gate.fence(run_id)
+                gate.reconcile_cancelled(run_id)
             raise
         except BaseException:
             # Pre-persist errors, torn persists, and validation failures: