Spark 3.5: Adapt PlanningBenchmark for DVs

[aokolnychyi]

This PR adapts our PlanningBenchmark for DVs.

Benchmark                                                                          (type)  Mode  Cnt            Score              Error   Units

PlanningBenchmark.distributedPlanningWithMinMaxFilter                           partition    ss    5            1.415 ±            0.553    s/op
PlanningBenchmark.distributedPlanningWithMinMaxFilter                                file    ss    5            5.000 ±            1.021    s/op
PlanningBenchmark.distributedPlanningWithMinMaxFilter                                  dv    ss    5            2.312 ±            0.307    s/op

PlanningBenchmark.distributedPlanningWithoutFilter                              partition    ss    5            2.147 ±            1.272    s/op
PlanningBenchmark.distributedPlanningWithoutFilter                                   file    ss    5            6.452 ±            2.327    s/op
PlanningBenchmark.distributedPlanningWithoutFilter                                     dv    ss    5            2.863 ±            0.590    s/op

PlanningBenchmark.distributedPlanningWithoutFilterWithStats                     partition    ss    5           49.325 ±            7.451    s/op
PlanningBenchmark.distributedPlanningWithoutFilterWithStats                          file    ss    5           54.895 ±           12.604    s/op
PlanningBenchmark.distributedPlanningWithoutFilterWithStats                            dv    ss    5           50.448 ±            3.760    s/op

PlanningBenchmark.localPlanningWithMinMaxFilter                                 partition    ss    5            1.313 ±            0.645    s/op
PlanningBenchmark.localPlanningWithMinMaxFilter                                      file    ss    5            3.546 ±            2.172    s/op
PlanningBenchmark.localPlanningWithMinMaxFilter                                        dv    ss    5            2.421 ±            1.126    s/op

PlanningBenchmark.localPlanningWithoutFilter                                    partition    ss    5            3.163 ±            0.283    s/op
PlanningBenchmark.localPlanningWithoutFilter                                         file    ss    5            5.455 ±            0.859    s/op
PlanningBenchmark.localPlanningWithoutFilter                                           dv    ss    5            4.296 ±            0.415    s/op

PlanningBenchmark.localPlanningWithoutFilterWithStats                           partition    ss    5           10.992 ±            3.095    s/op
PlanningBenchmark.localPlanningWithoutFilterWithStats                                file    ss    5           19.860 ±           23.286    s/op
PlanningBenchmark.localPlanningWithoutFilterWithStats                                  dv    ss    5           11.543 ±            6.390    s/op

The benchmark is set up so that there is a lot of more delete metadata to read with file-scoped deletes and DVs. Therefore, they are slower. DVs are faster than file-scoped deletes because there is less garbage and fewer GC operations.

This work is part of #11122.

[jbonofre]

LGTM, thanks ! I will run benchmark on my machine (curiosity 😄 ).

[aokolnychyi]

Thanks, @jbonofre @nastra!