test: the periodic test no longer fails across the top of an hour

[giraffesyo]

Summary

TestPeriodicJobsRunOnceAcrossLeaders failed the canary CI run of #42 on Postgres 17 (RunOnStart jobs = 2, want 1). It fails whenever it runs across the top of an hour; no library code changes.

The test checks that RunOnStart inserts one job, using a periodic job on Every(time.Hour). An hourly schedule also has a regular slot at the top of each hour, and the failing package run ended at 02:00:04 UTC after 30 seconds, so the test was running at 02:00:00 and the regular slot inserted a second job. Both jobs were correct.

The job now uses a schedule with no slots, so RunOnStart is the only thing that can insert it.

Testing

• go test -race -count=3 -run TestPeriodicJobsRunOnceAcrossLeaders . passes against Postgres 17.
• golangci-lint: 0 issues.
• Not reproduced on demand, since that needs the clock at the top of an hour; the diagnosis rests on the failure's timestamp. No other test starts a client with a schedule that depends on the time of day.

Only a test changed, so no hopperbench results.

Checklist

• make check passes — lint and the changed test run locally; the full suite runs in CI here
• No cryptography was added (tests run with GODEBUG=fips140=only)
• New or changed SQL in the claim, finalize, rescue or leader paths has a concurrency or chaos test — no SQL changed
• Behavior changes are reflected in docs/PLAN.md — none