It would be useful when debugging "RW is falling behind" alerts and high resharding to have a graph of remote write's latencies, either as a heatmap via the buckets or histogram_quantile to show p95/p99. If the query is somewhat complex we should consider including, or at least documenting, a recording rule.