Add query-rate windows (QPS) to /status recent_queries

[gaurav]

#231 adds recent_queries with mean and end-to-end p50/p95/p99 latency, but deliberately does not carry over the rate-tracking from #253:

• queries_last_10s / _60s / _300s and the matching queries_per_second_*
• inter_arrival_ms (mean/median/min/max/p95)
• history_span_seconds, time_since_last_query_seconds

Reason for deferring: percentiles need only durations (which recent_query_times already stores), but rate windows need a timestamp per query, i.e. reverting recent_queries to #253's (timestamp, duration) query_log design. That is a real design change rather than a few lines, so it was left out of #231.

Value: it directly answers "is Solr slow because it is overloaded, or because something is wrong with it?" — currently approximated via solr_metrics.host.system_load_average / system_cpu_load and by sampling query_handler.requests twice.

If implemented, note #268's finding that perf_counter_ns is monotonic-since-boot and produced a nonsense time_since_last_query_seconds: 1.7e9 in production — use wall-clock time.time() for the timestamps.