fix: keep automations running when someone renames a session group or edits sharing (#201434)

Root cause: src/config/sessions/session-delivery-generation.ts:192 treats the unkeyed `{ all: true, scope: "sessions" }` session-list notification as a stored-row invalidation, so one sidebar group rename or sharing edit retires the delivery generation of every in-flight run on the Gateway.

Related: #185085, #186361

## What Problem This Solves

A scheduled automation, a restart-resumed reply or another background session turn can fail partway through with `Session delivery generation is unavailable; retry after session storage is ready.` when anyone renames or deletes a session group or edits session sharing, even in an unrelated chat. The run ends as an error and its reply is lost.

## User Impact

User impact: renaming or deleting a session group, or changing who a session is shared with, no longer kills automations and background turns that are running in other sessions.

## Why This Change Was Made

Group and sharing handlers publish an unkeyed session-list notification after their storage writers have already published the exact per-session facts. The sharing-preparation owner already treats that plain notification as harmless once its facts are retained (d53921ac1424, "shared-session runs lose access after session list updates"). The delivery-generation lease is the other consumer of the same notification and still invalidated on it. It now applies the same rule: a plain `sessions` notification is ignored once the lease has retained facts, and every later check still reads the retained source and current facts. Notifications that carry `factsInvalidated` or a `runtimeSession` still invalidate, as before.

The change is in the lease, not in each publisher, so group rename, group delete and sharing edits are all covered without changing what clients receive.

## Evidence

- Production (Team Gateway, 14 days, read-only journal): 28 lines with this error. All of them come from session-event runs (cron `lane=cron-nested` / `source=cron`: 14, restart-recovery turns: 13, one webchat follow-up). The first attempt had been running for 19 s to 91 min before it failed, and several sessions failed in the same second, which points to one Gateway-wide invalidation rather than a per-session write. No exec-completion, subagent or channel-send callers.
- Isolated Gateway, deterministic local Responses fixture, cron job bound to a session whose turn takes 30 s, `sessions.groups.rename` called mid-run. Observed lines: https://github.com/openclaw/openclaw/pull/201434#issuecomment-6990134148
  - main 3bc20e28b022 (source and dist): every run that overlapped a rename failed with `Session target preparation outlived its attempt | admitted run authority is no longer active | Session delivery generation is unavailable` and `session event execution failed source=cron`, the same lines Team logs.
  - this branch ff77fa145849 (source and dist): all cron runs finish `ok` with the same renames.
  - control, main without the rename: the cron run finishes `ok`.
- Regression: `session-delivery-generation.native.test.ts` emits the unkeyed `sessions` notification on a prepared lease; red on main, green here. The same test checks that a `factsInvalidated` notification still retires the lease.
- Adjacent suites pass: delivery-generation diagnostics/incarnation/registry, sessions-sharing, sessions-groups, session-sharing-publication integration, sessions-rewind local-workspace handoff, sessions-mutations catalog-queue.

## Merge danger

Two-way door: one guard in the lease, no stored state, schema or config. If it were wrong, a run could keep a delivery generation across a group or sharing edit, but every effect still re-reads the retained source and the current row facts (membership, category, entry and permission changes are published per session and keep invalidating). Blast radius: in-flight session-event runs (cron, restart resume, exec completions) during session-list edits.

Thanks @jorge-monalee and @hvhoon for the detailed reports of runs dying with this error.

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