Several projection improvements

[ritchie46]

The lazy computation graph has schema for every node. This PR makes sure we actually use it and therefore save a lot column name lookups.

Test run

use polars::prelude::*;

fn main() -> Result<()> {
    let cols = (0..100000)
        .map(|i| Series::new(&format!("{i}"), [i]))
        .collect();
    let df = DataFrame::new_no_checks(cols);
    dbg!("created df");
    let names = df.get_column_names();
    dbg!("created names");

    let exprs: Vec<_> = (0..1000).map(|i| col(names[names.len() - 1 - i])).collect();

    for i in 0..10 {
        dbg!(i);
        df.clone().lazy().select(exprs.clone()).collect().unwrap();
    }

    Ok(())
}

Master

         38.066,31 msec task-clock                #   10,185 CPUs utilized          
             9.725      context-switches          #  255,475 /sec                   
                85      cpu-migrations            #    2,233 /sec                   
            17.976      page-faults               #  472,229 /sec                   
   117.932.283.930      cycles                    #    3,098 GHz                    
    43.582.787.522      instructions              #    0,37  insn per cycle         
    12.368.895.419      branches                  #  324,930 M/sec                  
         1.413.148      branch-misses             #    0,01% of all branches        

       3,737638335 seconds time elapsed

      37,864670000 seconds user
       0,181020000 seconds sys

This PR

            551,74 msec task-clock                #    1,040 CPUs utilized          
               569      context-switches          #    1,031 K/sec                  
                65      cpu-migrations            #  117,810 /sec                   
            17.982      page-faults               #   32,592 K/sec                  
     1.705.472.128      cycles                    #    3,091 GHz                    
     2.583.767.201      instructions              #    1,51  insn per cycle         
       497.490.863      branches                  #  901,681 M/sec                  
           436.293      branch-misses             #    0,09% of all branches        

       0,530435470 seconds time elapsed

       0,496682000 seconds user
       0,059751000 seconds sys

[ritchie46]

@ghuls