restart: stop hydration history flaking on worker-0 discovery

[antiguru]

The hydration-history-survives-restart workflow searched for a materialized view whose persist-sink worker sits off worker 0, so that the view's durable finish can be told apart from worker 0's earlier compute-only finish. That search is a dice roll: the sink worker is hash(sink_id) % workers, and a fast snapshot write can collapse both workers' stamps into one logging batch. When no trial separated the workers, the fixture aborted the run, which surfaced in CI as a flake with no product defect behind it.

builtins.AssertionError: test fixture did not produce an MV with its persist-sink worker off worker 0: [('hydration_history_mv_a', ...), ('hydration_history_mv_b', ...)]

Discovery now decides only which trial reports a regression, not whether the test passes. The loop still creates fresh views to roll the sink worker and the snapshot timing, and it still prefers a separating trial. Every fully hydrated trial is then checked against the all-worker maximum, which is the property under test whichever worker holds that maximum. A run where no trial separates the workers therefore still checks the property, it only loses the ability to tell a worker-0-only implementation apart on its own, and that query shape is pinned deterministically by collect_requires_every_worker.

The change is confined to test/restart/mzcompose.py.

🤖 Generated with Claude Code

[antiguru]

Thanks for the review!