Get aggregated fleet observability stats

Use when: caller wants a one-shot summary (request count, error rate, latency, cost, quality) across all of the principal’s gates

Headline numbers across every gate the caller owns — total requests, total cost, average latency, success rate, active gate count, total tokens, failure count. Scope to a project with projectId; defaults to a 30-day window (days, capped at 90).