test/migrate-after.test.ts's mirror cases pass every time on a loaded check machine #4622

The three mirror cases in test/migrate-after.test.ts now wait for the mirror push to end instead of running a 90 s Date.now() poll with 25 ms naps. Only the test file changed. I ran bun test on that file plus a check-only run, and haven't run scripts/flake-hunt.ts or the forge's full check.

• Event: the test hands startMirrorOut a thin wrapper around pushToMirrorTarget. When a push ends, whether it succeeded or threw, the wrapper wakes whoever is waiting.
• Waiting: settledTarget checks the mirror target row and, if the condition isn't met yet, sleeps until the next push ends and checks again. There is no fixed sleep and no raw deadline. The test's own timeout is the only bound, so I didn't need the #4431 budget helper.
• Not changed: src/ is untouched, and the assertions are the same.

I couldn't reproduce the original failures, so I haven't proved the root cause. I think they came from the poll loop's 90 s deadline running out on a loaded machine.

How it was run: bun test test/migrate-after.test.ts passed on its own (9 pass, 0 fail). I then ran 8 copies at once with --randomize, three rounds in a row, with no failures. That's 24 runs, which meets the "8 copies at once" option in the issue, not the flake-hunt.ts option.

bun x tsc --noEmit couldn't check the types because the bun type definitions weren't found in this sandbox, so the forge's check will be the first type check of this change.

Closes #4577

The three mirror cases in test/migrate-after.test.ts now wait for the mirror push to end instead of running a 90 s Date.now() poll with 25 ms naps. Only the test file changed. I ran bun test on that file plus a check-only run, and haven't run scripts/flake-hunt.ts or the forge's full check.

• Event: the test hands startMirrorOut a thin wrapper around pushToMirrorTarget. When a push ends, whether it succeeded or threw, the wrapper wakes whoever is waiting.
• Waiting: settledTarget checks the mirror target row and, if the condition isn't met yet, sleeps until the next push ends and checks again. There is no fixed sleep and no raw deadline. The test's own timeout is the only bound, so I didn't need the #4431 budget helper.
• Not changed: src/ is untouched, and the assertions are the same.

I couldn't reproduce the original failures, so I haven't proved the root cause. I think they came from the poll loop's 90 s deadline running out on a loaded machine.

How it was run: bun test test/migrate-after.test.ts passed on its own (9 pass, 0 fail). I then ran 8 copies at once with --randomize, three rounds in a row, with no failures. That's 24 runs, which meets the "8 copies at once" option in the issue, not the flake-hunt.ts option.

bun x tsc --noEmit couldn't check the types because the bun type definitions weren't found in this sandbox, so the forge's check will be the first type check of this change.