[FIX] Drive the insertion deadlines from an injectable clock

[jhampton]

This is the injectable-clock fix you sketched as option 2 on #82, after the reruns on #68 pushed the flake from occasional to constant. The clipboard reader's snapshot deadline and the copy wait now run on an InsertionClock seam in Dependencies, continuous in production and driven by hand in the tests, so the whole family holds or advances time itself: the happy-path tests arm deadlines that can never elapse, and the timeout tests force the deadline instead of racing a real timer against the runner, which makes the assertion "the deadline was honoured" rather than "the machine was fast enough". That also retires the 2-second busy-skip ceiling whose own comment recorded a CI failure at 50ms, and the 1-second scheduling waits I flagged on #68 become generous event budgets that only spend themselves on a genuinely broken run. I routed waitForCopy through the same clock deliberately: it is the same real-deadline shape in the same file and one seam serves both, which is a shade past what I promised on #82, so I would rather say it here than let the diff imply it. Production behaviour is unchanged, since the live clock is the same continuous clock the deadlines used before. The full suite is green locally on Apple Silicon at 433 passed with the four expected media-fixture skips, and the insertion class now finishes with no real sleeping at all. I used Claude Code for most of the writing under my direction and can explain every line.

Closes #82