4.0: grace period for bounded retention (W2)

[DanielCech]

Stacked on #23. Step 3 of the 4.0.0 plan.

Why

The hardest part of the JustFlip proof of concept was telling a real leak from an object that merely went away late: a store whose task finished after the timeout under simulator load looked exactly like a leak. Finding out meant rerunning the test alone with a longer timeout.

What's new

An object still alive at the timeout is watched for a grace period (3 s by default):

• Released in that time: a warning, and the test passes.

  App.InterestListStore was released after 3.2 sec, later than the 2 sec timeout. That's bounded retention, not a leak: something kept it alive for a while, e.g. a task, an animation or a delayed callback. If that's expected, raise the timeout.
  

• Still alive: a leak, as before. The message adds "It was watched for another 3 sec after that."

Set it with .deallocationGracePeriod(_:) or withDeallocationConfiguration; .zero turns it off. Passing checks are as fast as before. Only real leaks take the timeout plus the grace period to report.

Also fixed: durations in messages used Duration.formatted, which depends on the locale (on a Czech machine: "3 sec, 250,0 ms … 2,0 sec"). They now read the same everywhere: "400 ms", "2 sec", "3.2 sec".

Tests

• Swift Testing:
  • a late release is a warning;
  • a leak still fails after the grace period, and the message mentions it;
  • .zero turns the grace period off.
• XCTest: a late release passes as a non-strict expected failure.
• Core: the message wording and locale-independent durations.
• macOS (both trait configurations) and the iOS simulator pass.

🤖 Generated with Claude Code