Waits: a pr wait's repository pin made during the wait's first write is lost until the next check #118

From the concurrency-and-abuse review of #116 (#116 (comment)), test gap.

The step that saves a new pr wait's repository and host has no test for one case: when it happens while the new wait is still being written, the save is lost and done again at the next check. If the project's remote changes in that short gap (about one sweep), the wait follows the new remote.

Add a test for this case, and keep the save (or make the next check use the target judged at registration).

🤖 Generated with Claude Code