Ops-notification hourly ceiling — parsed limit + current window usage

Return the live state of every global ops-notification ceiling.

Diagnostic for the PR #519 finding-2 ceilings. The limits are env-tunable and prod has no shell access, so without an HTTP view of the parsed values a missed CD env-render hop would be invisible until a flood either went through or silently didn't. Also reports which backend is in play: redis means the ceilings are genuinely global; memory means Redis is unreachable and they have degraded to per-worker.

Counts only — no lead addresses, nothing PII-bearing. Read-only: calling this never charges the budget.

Shape: { "enterprise_lead": { "channel": "enterprise_lead", "backend": "redis", "windows": { "hour": {"limit": 10, "window_seconds": 3600, "used": 3, "remaining": 7, "reset_seconds": 1840, "digest_sent": false}, "day": {"limit": 20, "window_seconds": 86400, ...} } } }