Add scheduled Automations

[viannaanalyst]

What

Adds Automations: scheduled agent runs that execute while the app is open. Each automation is a recipe (project, harness, model, access mode, prompt, schedule) that owns one dedicated session, so context accumulates across runs.

• Schedules (local time) — every N hours, daily, weekdays, or weekly (a day of week); each with hour/minute where applicable, computed in the webview's timezone with a pure, tested module.
• Run lifecycle — claim-before-dispatch in SQLite (conditional UPDATE on the expected slot, single-winner across windows/reloads), missed slots recorded when the app was closed (no catch-up), and busy sessions skipped with a recorded reason.
• Dispatch — resolves the dedicated session (reuses it, unarchives before reuse, or creates it from the stored recipe and persists it), submits through onSubmit(..., { onSettled }), and records completed / failed / cancelled / skipped_busy plus counters. A 30-minute settlement guard closes runs whose onSubmit returns without settling, without charging the failure policy.
• Failure policy — pause after 1, 3, or 5 consecutive failures, or keep running; a success resets the counter, and a manual pause is preserved even if a run settles later.
• Surface — full-screen Automations view (list, next-run countdown, run history with errors and session links, run-now, pause/resume, edit, delete with confirmation) opened from a rail action below Kanban and from the View menu (in-app and native macOS). A dot marks failure-paused automations.
• Editor — title, prompt, project picker, harness/model/access pickers, schedule preset, failure policy.
• i18n — pt-BR for every new string, including dynamic schedule/countdown labels (translated through { key, vars } labels so placeholders resolve).

Dedicated-only v1 (no heartbeat mode), attended-only (no background execution), no natural-language stop conditions — all explicitly deferred in the spec.

Tests

• Web: npm run check:web — 234 files / 2420 tests passed, tsc --noEmit clean.
• Rust: npm run check:rust — fmt clean, clippy -D warnings clean, 317 tests passed.
• New coverage: schedule math (presets, boundaries, first-slot regression), failure policy + tick planning, store wrappers, Rust CRUD/claim/missed/result (conditional guards, cascade, counters), editor and surface static markup, rail entry.

Stacked PR

Based on feat/session-kanban (#2), which is based on feat/pr-write-actions (#1). Merge in order, or retarget to main once the earlier PRs land.

Pending before merge

• Manual dev-mode walkthrough: create an automation ~2 minutes ahead and watch it fire into the session transcript; verify busy-skip, missed-on-reopen, failure pause at the threshold, run-now, pause/resume, edit/delete (session kept), history, and both menu/rail entry points.