fix: stop empty automation checklists from recording failed runs and "did not start" notices (#202390)

Root cause: src/cron/service/timer-execution.ts:220 (and its twin in src/cron/session-run.ts:224) returns `status: "skipped"` when `skipIfScratchEmpty` finds an explicitly empty checklist, and every skipped run is a failure downstream: `resolveCronCompletionStatus` maps skipped to `completionStatus: "failed"`, run storage records it as failed, it counts toward skipped-run alerts, and the Gateway posts "did not start" for any skipped run without an admission-deferral reason.

Fixes #202350. Thanks @herosuncy for the clear report and repro.

## What Problem This Solves

A heartbeat-style automation with "Skip empty scratch" on and the default comments-only checklist skips every interval, as intended. But each skip is recorded as a failed run, and the owner conversation gets `Automation "heartbeat-main" did not start: Automation scratch is explicitly empty. Check automation history, resolve the cause, then run it again.` every 30 minutes. An idle heartbeat looks broken, and the only workaround was disabling it.

## User Impact

An empty checklist now completes quietly: run history shows `status: "ok"`, `completionStatus: "succeeded"` and the same summary `Automation scratch is explicitly empty`, the model is still not called, no "did not start" notice is posted, and it no longer counts toward skipped-run alerts. Real start failures (for example a model that is not configured) still record a failed run and still post the notice.

## Why This Change Was Made

The skip is a deliberate "nothing to do" decision, not a failed start. The cron outcome model has no non-failure skipped state, but it already has one for a run that completes with nothing to deliver: `status: "ok"` with a delivery suppression reason (the same outcome a heartbeat turn with nothing to report records). Both empty-scratch producers now return that outcome with the existing `"empty"` suppression reason, so completion, storage, failure alerts and the conversation notice all treat it as a quiet completion through their existing rules. The primary webhook already ignored skips that never started (#200032); its guard now ignores any non-error run that never started, so the empty-checklist summary is still not POSTed as output. No new status, field, wording or stored shape.

Behavior notes: `lastRunStatus` for these ticks is `ok` instead of `skipped`, and `consecutiveSkipped` no longer grows. A one-shot (`at`) job with `deleteAfterRun` and an empty checklist is now deleted like any other succeeded one-shot instead of being kept disabled; it was already consumed and never replayed.

History: the skipped return came with the Doctor heartbeat migration (ec5d369a6f66, #179387). It predates the "did not start" notice, and #194276 only kept the empty-scratch notice as an untouched control while fixing active-hours/busy deferrals, so recording it as a failure was never a product decision.

## Evidence

Real Gateway from source (`ocx src`), isolated state, deterministic local Responses provider that logs every model request, and a process-local Gateway clock stepped past each due time (the system clock is untouched). Job from the issue: every 30 min, `sessionTarget: session:agent:main:main`, `skipIfScratchEmpty: true`, scratch = the issue's comments-only template.

| | main e99e8482f8a8 | this PR 0519148ecf96 |
| --- | --- | --- |
| 2 scheduled ticks of heartbeat-main | `skipped` / `failed` x2 | `ok` / `succeeded` x2, same summary |
| job state | `consecutiveSkipped: 2` | `consecutiveSkipped: 0` |
| "did not start" notices in the session | 0 -> 2 | 0 -> 0 |
| model requests during the ticks | 1 -> 1 | 1 -> 1 |
| control: checklist with a real item | runs the turn (`ok`/`succeeded`, model called) | same |
| control: model override on an unconfigured provider | `error`/`failed` + "did not start" notice | same |

Observed output: https://github.com/openclaw/openclaw/pull/202390#issuecomment-7035458979

- `src/cron/service.runs-one-shot-main-job-disables-it.test.ts`: the existing scratch-empty one-shot case now asserts `ok`/`succeeded` history and no skip counter; it fails on main and passes here.
- `src/cron/service.trigger-eval.test.ts`: the existing webhook case still proves no POST for the empty checklist.
- Production: +7/-4 lines in three files. Tests: +9/-4 (the one regression).

Merge danger: two-way door. Nothing is migrated; a revert restores skipped/failed recording for later ticks, and rows written meanwhile stay `ok`/`succeeded`. Blast radius: automations with `skipIfScratchEmpty` and an empty checklist (run history status, completion, notices, skip alerts, `deleteAfterRun` one-shots).

Co-authored-by: Ayaan Zaidi <hi@obviy.us>