Clean up newly created task rows after schedule creation failure
When schedule creation rejects or returns no usable ID, a newly inserted task row currently remains even though creation failed. Attempt to delete only that new row before rethrowing the original failure. A cleanup failure is logged without replacing the scheduling error.
This is the separate cleanup slice of original #291; the cron parser fix is already released in #1100 and unchanged. Preserve current error identity/text instead of the old patch's new prefix. Successful results, time zone forwarding, auth, ownership, plan/cadence and worker behavior remain unchanged. Existing isolated tests mock the newly imported database helper.
This is best-effort local cleanup, not a transaction with Trigger. A network rejection/missing response ID does not prove no remote schedule was created; no unknown schedule is located/deleted. Initial insert/plan failures and failures saving the returned schedule association retain existing behavior. No historical repair, migration, new worker/job, dependency, billing or model change.
Validation: baseline 249 tests/37 files; new red fixture four failed/five passed; all nine cleanup tests green. Broader tasks/MCP/plans/scheduler/contracts suite: 483 passed/61 files. Lint and changed formatting pass. Whole-project TypeScript retains the same 207 pre-existing diagnostics byte-for-byte. Independent review at the exact head: no actionable Spec or Standards findings. Fresh CI: 7,372 tests passed/15 skipped; 1,301 files passed/13 skipped; response-contract type check, lint, format, compatibility and Vercel preview all passed. No live database delete, provider schedule/job, customer action, paid operation or manual deployment was performed.