Clearance throughput benchmark

Drives a configurable load against the live hosted Vault — N real /request-clearance calls in batches of M every interval, with live latency aggregation. Default is 1000 requests in batches of 20 per second (~50 seconds total). Every clearance is signed end-to-end and lands in the demo tenant's ledger; the page measures only the round-trip time to a verdict — no canned tool stubs are executed after.

Note: at 20 req/s the demo tenant's judge is doing ~20 real LLM evaluations per second. Bedrock Nova p50 is typically 150-400ms; rate limits or judge concurrency caps may surface as rate_limited in the outcomes panel.