Document agg pushdown on ORC file limitation [skip ci]

[amahussein]

closes #4950

Signed-off-by: Ahmed Hussein (amahussein) a@ahussein.me

• document the expected failure caused by Spark-3.3.0 assumption that all ORC files must have file statistics.

• document the usage of new spark-3.3.0 feature introduced in SPARK-34960

  • the limitations with RAPIDS while enabling this feature
  • how to avoid the SparkException

[revans2]

Can we say that the RAPIDS Accelerator does not support. Because technically CUDF does, just not for chunked writes, which is what we are doing. Or perhaps just say that we are working with CUDF to support this feature and you can track it here...

[amahussein]

I made the changes.

BTW, There was another CUDF issue 10075 which reports that statistics are missing without being specific to the mode. I commented on the issues to verify whether they are dup or that Cudf does not support the file statistics at all.

[amahussein]

build

[amahussein]

build

[amahussein]

build

[amahussein]

build

[sameerz]

build

[razajafri]

build

[amahussein]

build

[razajafri]

The PR has [skip ci] in the title. Why do you expect it to run ci?

[amahussein]

The PR has [skip ci] in the title. Why do you expect it to run ci?

I could not build previous PRs couple of weeks ago.
I wanted to get the authentication fixed. Skipping CI is fine so that I won't consume resources just for testing my authentication.