feat: accelerate Iceberg RewriteDataFiles reads via Comet native scan

[jordepic]

Which issue does this PR close?

Closes #4250.

Rationale for this change

A large number of query resources are devoted across the industry to rewriting data files using spark procedures for iceberg tables. Using native code here where possible can significantly speed up this process!

What changes are included in this PR?

Detect spark scans (SparkStagedScan) that are created during the RewriteDataFilesSparkAction and replace them with comet scans. Extract their associated tasks and pass in the lack of filter (see SparkStagedScan line 50 in the apache iceberg project).

Note that some things are NOT included in this PR:
Datafusion-comet does not yet support writing iceberg data. Once it does, we can "comet-ify" this whole pipeline!

How are these changes tested?

We write two tests to inspect the spark plan associated with rewriting data files and ensure that the operators get replaced. Before this change is merged I can also try to run it locally and pick up some benchmarks for table compactions (on tables that are only data files, and those with delete files associated).

I have also added tests to ensure that the compaction works, not just ensuring operator replacement, but also runtime compaction correctness on both bin packing and sorting!

[mbutrovich]

Spark 3.4 failed for a reflection access. IIRC we use an older version of Iceberg there, so signatures might have changed. We might need different reflection logic for older versions of Iceberg. I had to do that somewhere else in the reflection class, but can't recall what right now.

[mbutrovich]

Thanks @jordepic! I will take a proper pass through this today.