Add ChunkMeta to the index containing a `*rowRange`

[simonswine]

This would optimize how we read a block on disk. It also can help recovering row group segments from disk (see grafana/pyroscope#2076).