feat: Relaxed schema alignment for parquet file list read

[nameexhaustion]

Fixes #18568
Fixes #17254

• Support reading from a list of parquet files that have the same set of columns, but not necessarily in the same order

  • The dtype of every column must be equal

• When the files have differing columns:

  • If there are extra columns in files following the first file, those columns are silently ignored (not read)
    • Let me know if we want to error here instead
  • If there are columns in the first file missing from the files after it, an error will be raised. Later I will make a PR to introduce an option to instead project NULL rows for this case.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Please upload report for BASE (main@77b1e42). Learn more about missing BASE report.
Report is 12 commits behind head on main.

Additional details and impacted files

@@           Coverage Diff           @@
##             main   #18803   +/-   ##
=======================================
  Coverage        ?   79.86%           
=======================================
  Files           ?     1518           
  Lines           ?   205582           
  Branches        ?     2892           
=======================================
  Hits            ?   164180           
  Misses          ?    40854           
  Partials        ?      548           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[kszlim]

Does:

The dtype of every column must be equal

Imply no upcasting if you have f32 in one file and f64 in another for one column?

[ion-elgreco]

Does:

The dtype of every column must be equal

Imply no upcasting if you have f32 in one file and f64 in another for one column?

This shouldn't be enforced otherwise that would mean type widening won't be possible.

[ion-elgreco]

Any chance you will allow to pass a schema that should be used instead of fetching it from the metadata of the first file?

[coastalwhite]

Any chance you will allow to pass a schema that should be used instead of fetching it from the metadata of the first file?

This is tracked in #15111.