test: add tmux-driven TUI e2e scenarios for Monitor - #4086

 apps/kimi-code/test/e2e/
~├── tui-sandbox.ts               # 新增 monitor（可带 timeout / persistent）、foreground、stopMonitor、
~│                                #   waitForMonitor、monitorOutput 步骤；运行时按描述查找 monitor 的 task id；
~│                                #   startFakeService（模拟远程服务）；scenario 可传环境变量，
~│                                #   上下文提供 workDir 和 relaunch（用 --continue 重启同一会话）
+└── tui-monitor.e2e.test.ts      # 16 个 Monitor 场景

KIMI_E2E=1 pnpm --filter @moonshot-ai/kimi-code exec vitest run test/e2e/tui-monitor.e2e.test.ts

Requirement or Bug

Stacked on #4085. Adds end-to-end tests that drive the real TUI for the Monitor tool. It reuses the tmux harness from #4069; until #4069 merges, this PR's diff also contains those two files. Once #4069 lands, main will be merged in and the diff will shrink to this PR's own changes.

Bug Reproduction Steps

N/A

Root Cause

N/A

Code Changes

 apps/kimi-code/test/e2e/
~├── tui-sandbox.ts               # adds monitor (with timeout / persistent), foreground, stopMonitor,
~│                                #   waitForMonitor and monitorOutput steps; a monitor's task id is looked up
~│                                #   by description at run time; startFakeService (a stand-in remote service);
~│                                #   scenario takes env vars, and its context has workDir and relaunch
~│                                #   (restart into the same session with --continue)
+└── tui-monitor.e2e.test.ts      # 16 Monitor scenarios

Scenarios (the flag is turned on with KIMI_CODE_EXPERIMENTAL_MONITOR=1):

• Basics: an event wakes an idle agent; an event that arrives mid-turn reaches the next step; WaitFor with a monitor's task_id returns when it prints; WaitFor without a task_id waits for the build and is not ended by monitor output; a flooding monitor is stopped and the model hears why.
• Remote service status (background tasks + WaitFor cannot do this: they report once, when the command ends): every status change wakes the agent separately; an unreachable service wakes the agent; once the service is healthy the agent stops the monitor with TaskStop, and no events follow.
• From real use cases (Claude Code docs, blog posts, GitHub issues):
  • dev-server log: one wake per new error, and lines already delivered are not sent again
  • lines printed while an event is undelivered arrive together as one batch after it is consumed
  • a CI watch that exits: its last lines arrive before the failure notice, and stderr stays out of events
  • a burst larger than one batch (50 lines): the newest lines are kept, <event omitted="10">, the TUI shows the omitted count, and TaskOutput has the full log
  • hostile log lines (</event></notification>…, &, an overlong line) appear only as data inside the event
  • at its timeout a monitor sends one notice and its process is killed; it can then be re-armed with persistent
  • WaitFor on one monitor is not ended by another monitor's event
  • quitting kills the monitor process; resuming with --continue replays the event rows and starts no new turn

Behavior Changes and Affected Users

None — tests only. The suite is opt-in like the existing e2e tests: it runs only with KIMI_E2E=1 and tmux on PATH, and skips otherwise, so default vitest runs and CI are unchanged.

KIMI_E2E=1 pnpm --filter @moonshot-ai/kimi-code exec vitest run test/e2e/tui-monitor.e2e.test.ts

Checklist

• I have read the CONTRIBUTING document.
• I have linked a related issue (external PRs: issue must have a maintainer's /approve).
• I have added tests that prove my feature works.
• The behavior-change table above is complete, and every removed behavior or flipped default is named in the changeset and either has an escape hatch or was explicitly approved by a maintainer in this PR.
• Ran gen-changesets skill, or this PR needs no changeset.
• Ran gen-docs skill, or this PR needs no doc update.