Update speedup factors for qualification tool in Dataproc 2.1 environments

[cindyyuanjiang]

We reran NDS benchmarks in Dataproc 2.1 environments with latest versions, and updated the speedup factors for Dataproc 2.1 T4 and L4 env.

Fixes #508

[amahussein]

I cannot see which tests failed.
@cindyyuanjiang can you please rerun the unit tests locally for spark 341 to see what is going on.

[amahussein]

I guess the ML ops added to the t4/l4 were copy/paste from the operatorsScore.csv. Was that done on purpose? AFAIK, the ML-ops are not parts of the benchmarks.
I am concerned that we update the file skipping some operators that were not actually part of the evaluation.

• If the ML-functions is added to the file, then teveryone is assuming the speedups are up-to-date.
• If the speedups for ML-functions are not measured on t4/l4, then they should not be added to the file.

Finally, we should consider upgrading the default build to >= Spark-330 since:

1. speedups are based on dataproc-2.1 which runs Spark330.
2. user-tools downloads spark-3.3.1 by default.

Prefer a second opinion from @nartal1

[nartal1]

I second that speedups for ML ops should not be added to t4/l4. It would imply that we got these numbers by testing and running on t4/l4. The speedups were obtained from databricks-aws.

[cindyyuanjiang]

@amahussein @nartal1
Thank you very much for pointing this out! The ML op scores were generated from the script automatically. I have removed them from t4/l4 operator score files.