fix(queue): harden drain/abort/timeout race handling

- reject new lane enqueues once gateway drain begins
- always reset lane draining state and isolate onWait callback failures
- persist per-session abort cutoff and skip stale queued messages
- avoid false 600s agentTurn timeout in isolated cron jobs

Fixes #27407
Fixes #27332
Fixes #27427

Co-authored-by: Kevin Shenghui <shenghuikevin@github.com>
Co-authored-by: zjmy <zhangjunmengyang@gmail.com>
Co-authored-by: suko <miha.sukic@gmail.com>

src/cli/gateway-cli/run-loop.test.ts

@@ -9,6 +9,7 @@ const consumeGatewaySigusr1RestartAuthorization = vi.fn(() => true);
 const isGatewaySigusr1RestartExternallyAllowed = vi.fn(() => false);
 const markGatewaySigusr1RestartHandled = vi.fn();
 const getActiveTaskCount = vi.fn(() => 0);
+const markGatewayDraining = vi.fn();
 const waitForActiveTasks = vi.fn(async (_timeoutMs: number) => ({ drained: true }));
 const resetAllLanes = vi.fn();
 const restartGatewayProcessWithFreshPid = vi.fn<
@@ -37,6 +38,7 @@ vi.mock("../../infra/process-respawn.js", () => ({
 
 vi.mock("../../process/command-queue.js", () => ({
   getActiveTaskCount: () => getActiveTaskCount(),
+  markGatewayDraining: () => markGatewayDraining(),
   waitForActiveTasks: (timeoutMs: number) => waitForActiveTasks(timeoutMs),
   resetAllLanes: () => resetAllLanes(),
 }));
@@ -213,6 +215,7 @@ describe("runGatewayLoop", () => {
       await new Promise<void>((resolve) => setImmediate(resolve));
 
       expect(waitForActiveTasks).toHaveBeenCalledWith(30_000);
+      expect(markGatewayDraining).toHaveBeenCalledTimes(1);
       expect(gatewayLog.warn).toHaveBeenCalledWith(DRAIN_TIMEOUT_LOG);
       expect(closeFirst).toHaveBeenCalledWith({
         reason: "gateway restarting",
@@ -229,6 +232,7 @@ describe("runGatewayLoop", () => {
         restartExpectedMs: 1500,
       });
       expect(markGatewaySigusr1RestartHandled).toHaveBeenCalledTimes(2);
+      expect(markGatewayDraining).toHaveBeenCalledTimes(2);
       expect(resetAllLanes).toHaveBeenCalledTimes(2);
       expect(acquireGatewayLock).toHaveBeenCalledTimes(3);
     });

src/cli/gateway-cli/run-loop.ts

@@ -9,6 +9,7 @@ import {
 import { createSubsystemLogger } from "../../logging/subsystem.js";
 import {
   getActiveTaskCount,
+  markGatewayDraining,
   resetAllLanes,
   waitForActiveTasks,
 } from "../../process/command-queue.js";
@@ -111,6 +112,9 @@ export async function runGatewayLoop(params: {
         // On restart, wait for in-flight agent turns to finish before
         // tearing down the server so buffered messages are delivered.
         if (isRestart) {
+          // Reject new enqueues immediately during the drain window so
+          // sessions get an explicit restart error instead of silent task loss.
+          markGatewayDraining();
           const activeTasks = getActiveTaskCount();
           if (activeTasks > 0) {
             gatewayLog.info(