improve agg_list performance of chunked numerical data

[ritchie46]

Delaying a full rechunk can improve performance of many query, but some are slower.

This improves the following query by ~3x.

fn main() -> Result<()> {
    let q = LazyCsvReader::new("/home/ritchie46/code/db-benchmark/data/G1_1e7_1e2_0_0.csv".into())
        .with_rechunk(false)
        .finish()?;

    let out = q
        .groupby([col("id2"), col("id4")])
        .agg([(pearson_corr(col("v1"), col("v2")).pow(2.0)).alias("r2")])
        .collect()?;
    Ok(())
}

before

          6.685,79 msec task-clock                #    2,124 CPUs utilized          
             4.747      context-switches          #  710,013 /sec                   
               218      cpu-migrations            #   32,606 /sec                   
           173.797      page-faults               #   25,995 K/sec                  
    21.435.123.551      cycles                    #    3,206 GHz                    
    38.153.190.822      instructions              #    1,78  insn per cycle         
     5.010.104.768      branches                  #  749,366 M/sec                  
        24.141.092      branch-misses             #    0,48% of all branches        

       3,148083894 seconds time elapsed

       6,231878000 seconds user
       0,469579000 seconds sys

after

          5.024,99 msec task-clock                #    4,879 CPUs utilized          
             4.305      context-switches          #  856,718 /sec                   
               118      cpu-migrations            #   23,483 /sec                   
           212.864      page-faults               #   42,361 K/sec                  
    14.441.547.675      cycles                    #    2,874 GHz                    
    21.711.917.241      instructions              #    1,50  insn per cycle         
     4.031.158.541      branches                  #  802,222 M/sec                  
        21.774.943      branch-misses             #    0,54% of all branches        

       1,029876114 seconds time elapsed

       4,531616000 seconds user
       0,502194000 seconds sys

[codecov-commenter]

Codecov Report

Merging #3351 (d4a7cf3) into master (89ebc59) will decrease coverage by 1.44%.
The diff coverage is 88.39%.

@@            Coverage Diff             @@
##           master    #3351      +/-   ##
==========================================
- Coverage   77.56%   76.11%   -1.45%     
==========================================
  Files         390      393       +3     
  Lines       67948    68396     +448     
==========================================
- Hits        52703    52060     -643     
- Misses      15245    16336    +1091     

Impacted Files	Coverage Δ
polars/polars-arrow/src/bitmap/mutable.rs	100.00% <ø> (+22.22%)	⬆️
polars/polars-core/src/chunked_array/object/mod.rs	58.90% <ø> (ø)
polars/polars-io/src/csv_core/buffer.rs	78.21% <0.00%> (ø)
polars/polars-core/src/frame/row.rs	65.95% <14.28%> (-0.72%)	⬇️
polars/polars-core/src/frame/groupby/mod.rs	83.15% <72.22%> (-0.59%)	⬇️
...lars/polars-core/src/frame/groupby/aggregations.rs	77.38% <98.83%> (+0.39%)	⬆️
polars/polars-time/src/lib.rs	0.00% <0.00%> (-100.00%)	⬇️
polars/polars-utils/src/lib.rs	0.00% <0.00%> (-100.00%)	⬇️
polars/polars-time/src/windows/test.rs	0.00% <0.00%> (-100.00%)	⬇️
polars/polars-time/src/groupby/dynamic.rs	52.38% <0.00%> (-40.37%)	⬇️
... and 32 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 4e6fa11...d4a7cf3. Read the comment docs.