[benchmark] Fix stale throughput and duration docs in cluster README

[thswlsqls]

Purpose

Two README lines still describe the behavior from before 8815608 (#274), although that commit updated the neighbouring lines.

• Throughput/Core is rps / cpu, where rps reads Flink's numRecordsOutPerSecond (JobBenchmarkMetric#getRpsPerCore) — rows, not bytes. The README also names the column Throughput/Cores, while Benchmark prints Throughput/Core.
• Writing stops after row-num rows from queries.yaml, not after 30 minutes: that commit removed benchmark.metric.monitor.duration and left a sentence contradicting "Each SQL script will stop only after it produces enough number of rows" below it.

No linked issue: docs wording fix, no open issue on this topic.

Tests

• Docs-only: one file, two lines, no behavior change, so no test was added. mvn spotless:check -pl paimon-benchmark/paimon-cluster-benchmark passes.

[JingsongLi]

-1 It's pointless.