This eliminates the need for storing block data into temporary files on a single-node VictoriaMetrics
during heavy queries, which touch big number of time series over long time ranges.
This improves single-node VM performance on heavy queries by up to 2x.