fix(desktop): restore stable cron jobs position, compact strip, and balanced sidebar spacing

[ValentinSergief]

Problem

Two regressions in the Desktop sidebar after the sidebar rework (#42537 / #43147):

1. The running dot never pulses – the original heuristic keyed off next_run_at being within ~10s of now (nowMs - next < 300_000 && nowMs - next > 0). Origin's PR fix(cron): prevent recurring job re-fire on gateway crash/restart loop #3396 (at-most-once crash-safety) moved advance_next_run() to before run_job(), so next_run_at is always pre-advanced to the next future slot while a job runs. The heuristic never fires.

2. Triggered jobs jump to the top of the list, then drop back – trigger_job() persists next_run_at = now as the scheduler's "fire now" contract. During the seconds before the scheduler claims the job, the sidebar sorts it to the top, then drops it back to its real slot. Normally (scheduled runs) this never happens because the scheduler pre-advances before the frontend polls – the jump is unique to manual triggers.

Fix

Replace the client-side timing heuristic with an authoritative execution ledger, and stop serving the transient next_run_at = now to the UI.

• cron/executions.py – active_execution_start_times(): cross-process durable ledger of in-flight job starts, with a 30-min claim TTL staleness bound (health-check wedge: a wedged claim can't pin the dot forever).
• cron/jobs.py – list_jobs() / get_job() stamp authoritative is_running + active_run_started_at from the ledger. A display guard never serves a past next_run_at for a scheduled-but-idle job – the computed next slot is shown instead, so the sort key stays stable through a manual trigger.
• tui_gateway/server.py – _cron_sig folds the running set into the change signature, so the desktop learns a job went live immediately instead of waiting on jobs.json mtime.
• apps/desktop – the dot and "X ago" counter are driven by active_run_started_at (jobStartRef grace counter removed). Balanced search-bar spacing retained.

7 files, +287/-27. No new dependencies, no scheduler-contract changes – the scheduler still reads the persisted next_run_at = now and fires on the next tick.

Testing

1. Trigger a cron job from the sidebar – dot pulses immediately, "X ago" counts, and the job stays in its sorted position (no jump to top / drop back).
2. Let the run finish – dot stops, counter flips to "in X min" at the real next slot.
3. Health-check wedge: a claimed-but-wedged execution expires after 30 min instead of pinning the dot.
4. Idle state – no dot, stable position.
5. tests/cron/test_jobs.py + tests/cron/test_execution_ledger.py: 154 tests pass (4 new contract tests for the stamping display guard). ruff clean, tsc clean on both desktop configs.
6. Verified locally in the running app (rebuild + swap): dot pulses and clears, no position jump.

Fixes #43309

[tonydwb]

Code Review Summary

Verdict: Approved

UI polish fix for the desktop cron jobs sidebar section. Tightens the max-height of the cron jobs list and repositions it below the sessions section with better visual separation.

Looks Good

• max-h-72 reduced to max-h-28 prevents the cron strip from consuming excessive vertical space
• Cron jobs section now lives below sessions with its own shrink-0 container — more stable layout
• Running-state pulse animation improved with a proper ping effect
• pb-1 instead of pb-2 for better spacing balance
• No security concerns

---

Reviewed by Hermes Agent

[ValentinSergief]

Thanks for the review, @tonydwb. Pushed a spacing correction on top of this — nav pb-1→pb-0, search pb-1→pb-1.5 (the p-2 padding inside SidebarMenuButton was invisibly stacking). Both gaps now equal at 12px for visual spacing balance.

Otherwise unchanged. Ready for CI and maintainer review.

[ValentinSergief]

Verified compatible with #44630 (composer status stack). Sidebar spacing + cron dot work correctly alongside the new composer UI.

[ValentinSergief]

The running indicator is now driven by the execution ledger's active_run_started_at (stamped at dispatch in cron/executions.py, served through list_jobs()/get_job() in cron/jobs.py, folded into the _cron_sig change signature) rather than the transient next_run_at observation — a job that starts while the section is collapsed still shows its run when the section reopens, and the 30-min claim TTL bounds staleness. Contract tests in tests/cron/test_jobs.py cover the stamping.