WIP: Reduce sort memory usage v2 by @richox

[yjshen]

Which issue does this PR close?

Closes #.

Rationale for this change

Another trial to reduce memory usage during the sort.
Mainly authored by @richox

What changes are included in this PR?

Are there any user-facing changes?

[yjshen]

cargo run --release --features "mimalloc" --bin tpch -- benchmark datafusion --iterations 3 --path ../tpch-parquet/ --format parquet --query 1 --batch-size 4096

Without this PR:

Running benchmarks with the following options: DataFusionBenchmarkOpt { query: 1, debug: false, iterations: 3, partitions: 2, batch_size: 4096, path: "../../tpch-parquet/", file_format: "parquet", mem_table: false, output_path: None }
Query 1 iteration 0 took 2851.7 ms and returned 6001214 rows
Query 1 iteration 1 took 2817.7 ms and returned 6001214 rows
Query 1 iteration 2 took 2735.9 ms and returned 6001214 rows
Query 1 avg time: 2801.75 ms

With this PR:

    Finished release [optimized] target(s) in 0.07s
     Running `target/release/tpch benchmark datafusion --iterations 3 --path ../tpch-parquet/ --format parquet --query 1 --batch-size 4096`
Running benchmarks with the following options: DataFusionBenchmarkOpt { query: 1, debug: false, iterations: 3, partitions: 2, batch_size: 4096, path: "../tpch-parquet/", file_format: "parquet", mem_table: false, output_path: None }
Query 1 iteration 0 took 36483.0 ms and returned 6001214 rows
Query 1 iteration 1 took 36783.2 ms and returned 6001214 rows
Query 1 iteration 2 took 36400.4 ms and returned 6001214 rows
Query 1 avg time: 36555.52 ms

There appears to be a serious deterioration in performance 😅

[alamb]

This looks very cool -- thank you @yjshen and @richox for working on this

[yjshen]

I am closing it for now because of the performance regression.