SortSpillManager is used to manage how to write spill data file and read them back.
The spill data first need to be partitioned in a round-robin way. Each partition contains 1 file: input_chunks_file
The spill file consume a data chunk and serialize the chunk into a protobuf bytes.
Finally, spill file content will look like the below.
The file write pattern is append-only and the read pattern is sequential scan.
This can maximize the disk IO performance.