test: make the heartbeat recovery test independent of write speed

[cardmagic]

Summary

• EffectRecoveryTest#test_a_running_effect_keeps_its_process_heartbeat_fresh failed three times on MySQL (trilogy and mysql2) in the last 30 failed CI runs, with Expected: "processing", Actual: "completed".
• Cause, in the test: it sets process_alive_threshold = 0.1 and runs the recovery pass right after the 8th heartbeat. One slow heartbeat commit makes the stored heartbeat older than 0.1 s, so the pass retires the effect, which is the designed behavior for a stale owner (docs/effect-recovery.md). With the defaults (15 s interval, 60 s threshold), a heartbeat must stall for about 45 s.
• Fix, test only: the test waits until the stored heartbeat is more than one threshold past the claim, then runs the recovery pass with database_clock_now set half a threshold after that heartbeat. The heartbeat thread and the database writes stay real.
• Test first: a new case pauses 0.15 s after each heartbeat write. It failed with the CI message before the change and passes after it. Both cases fail with Timeout::Error when ProcessHeartbeat#start is disabled, so they still catch a missing heartbeat.

Checks

• bundle exec rake passes on SQLite (984 runs, 0 failures).
• The two cases passed 10 times in a row.

[greptile-apps]

Confidence Score: 5/5

[Low impact] The test-only change appears safe to merge.

Summary

Makes the heartbeat recovery test independent of database write speed.

• The heartbeat test checks recovery at a controlled time.

_{Reviews (1) · Last reviewed commit: "test: make the heartbeat recovery test i..." · Reviewed by Greptile}