Revert "Ensure PartitionedOutputOperator is run with fixed local distribution"

[arhimondr]

Description

This reverts commit 1c06202.

Additional local exchange increases CPU time.

Regression noticed during performance verification cycle by @lukasz-stec

Reproduced by running a following query on a TPCH x1000 dataset on a 4 nodes cluster (m5.8xlarge on AWS):

SELECT
  sum(length(l_comment) + length(s_comment))
FROM lineitem, supplier
WHERE l_suppkey = s_suppkey;

The additional local exchange causes a CPU time increase from ~2.09h to ~2.19h (a 5% increase).

The regression is caused by an increase of context switches and CPU migrations:

Before:

  2,091,016.93 msec task-clock                #   27.119 CPUs utilized          
         2,221,498      context-switches          #    0.001 M/sec                  
           658,519      cpu-migrations            #    0.315 K/sec                  
               920      page-faults               #    0.000 K/sec                  
 5,386,582,982,247      cycles                    #    2.576 GHz                    
 4,446,312,739,476      instructions              #    0.83  insn per cycle         
   680,314,613,026      branches                  #  325.351 M/sec                  
    22,854,687,505      branch-misses             #    3.36% of all branches        
 1,078,160,950,603      L1-dcache-loads           #  515.616 M/sec                  
     5,255,101,736      L1-dcache-load-misses     #    0.49% of all L1-dcache hits  
   <not supported>      LLC-loads                                                   
   <not supported>      LLC-load-misses   

After:

 2,390,431.42 msec task-clock                #   27.578 CPUs utilized          
         3,026,056      context-switches          #    0.001 M/sec                  
           940,979      cpu-migrations            #    0.394 K/sec                  
             4,322      page-faults               #    0.002 K/sec                  
 6,185,926,599,130      cycles                    #    2.588 GHz                    
 4,777,419,838,837      instructions              #    0.77  insn per cycle         
   733,708,556,808      branches                  #  306.936 M/sec                  
    24,900,478,611      branch-misses             #    3.39% of all branches        
 1,160,521,729,359      L1-dcache-loads           #  485.486 M/sec                  
     4,050,568,791      L1-dcache-load-misses     #    0.35% of all L1-dcache hits

What results in an observable regression in the number of instruction per cycle (0.83 insn per cycle vs 0.77 insn per cycle).

Additional context and related issues

#13834

Release notes

(X) This is not user-visible or docs only and no release notes are required.
( ) Release notes are required, please propose a release note for me.
( ) Release notes are required, with the following suggested text:

# Section
* Fix some things. ({issue}`issuenumber`)

[arhimondr]

CC: @linzebing @losipiuk