Adding parquet transcoding example

[mhaseeb123]

Description

This PR adds a new example parquet_io to libcudf/cpp/examples instrumenting reading and writing parquet files with different column encodings (same for all columns for now) and compressions to close #15344. The example maybe elaborated and/or evolved as needed. #15348 should be merged before this PR to get all CMake updates needed to successfully build and run this example.

Checklist

• I am familiar with the Contributing Guidelines.
• New or existing tests cover these changes.
• The documentation is up to date with these changes.

[GregoryKimball]

Thank you @mhaseeb123 for writing this example! It has made it easy for me to do a ton of testing with Kaggle data 🤩

There are a few benchmarking items that we might want to address in the example, maybe with a comment, maybe with code change.. up to you.

1. startup time on first read. I believe we do nvcomp and cufile loading on the first read, so it takes more like 3s instead of ~30-300 ms. We might want to pre-load the extra libraries or else leave the first read untimed.
2. RMM pool growth. I believe the example uses a pool and then this grows during the first read, adding to runtime. We might choose to set a default, accept a CLI parameter, or leave the first read untimed.
3. OS cache clearing. Currently the example will time the second read optimistically, at least if the file is in the OS cache. I'm not sure if written files are always in the OS cache or if they have to be read once first. We might want to investigate to see if cache clearing before the second read has any impact on read time.
4. Allowing optional size metadata to be written. We added page size metadata to the writer PARQUET-2261 Size Statistics #14000 and reader Use page statistics in Parquet reader #14973. Perhaps we could enable this option via a CLI parameter.

[mhaseeb123]

Thank you @GregoryKimball for the feedback. I am thinking of not timing the first read to avoid both RMM pool growth, and nvcomp and cufile pitfalls. I will add a CLI option to clear OS cache before second and subsequent reads plus another CLI option for optional size metadata soon.

[GregoryKimball]

Thank you @mhaseeb123 for discussing the example. I think we will address 1,2,3 above by untiming the first read, and then resolve 4 by adding a writer option setting the sizes metadata to OFF

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

[mhaseeb123]

/ok to test

[mhaseeb123]

Hi reviewers @vuule @vyasr. Just pinging you as the PR is now ready to be reviewed

[mhaseeb123]

/ok to test

[mhaseeb123]

/ok to test

[vuule]

few more suggestions, looks good otherwise 👍

[mhaseeb123]

/ok to test

[vuule]

approved; just one optional suggestion

[mhaseeb123]

/ok to test

[mhaseeb123]

/ok to test

[mhaseeb123]

/merge