RPC Endpoint Benchmarker

Times repeated calls against each endpoint and reports the median rather than the mean, since one slow round trip on a shared connection drags an average badly. Alongside speed it probes the capabilities that decide whether tools work at all: eth_getLogs, historical state, and debug_traceTransaction. That matters more than latency for most work, because a fast endpoint that refuses trace calls is useless for gas profiling however quick it is. A JSON-RPC error arrives as a 200 response carrying an error object, so checking the HTTP status alone would report every unsupported method as working — the response body is inspected instead. Chain ids are read and a mixed set is called out.