[AlxcBTW]

Root causes found while measuring CI health (night/ci-health/flakes.md):

• macOS real_window::the_worker_saves_a_clip_of_a_real_window fails on main (clips: the system did not start the capture in time -> no_window): the start of a SCStream was never answered within 10 s on a loaded runner. The window source now retries such a start (3 tries); the test's wait for an event is 60 s to cover it.
• observe::a_title_log_stops_growing_at_its_cap (Linux, 2110 > 2048): tokio's write_all returns before the write landed, so the next line measured a stale size. The write is flushed before the next measurement.

---

^{Need help on this PR? Tag @codesmith-bot with what you need.}