feat: Native CSV file list reading

[nameexhaustion]

This replaces the existing codepath that dispatches multi-CSV file reads to the Union executor.

[codspeed-hq]

CodSpeed Performance Report

Merging #16180 will degrade performances by 22.39%

_{Comparing nameexhaustion:csv-file-list (dfc3bdb) with main (993cc99)}

Summary

❌ 1 regressions
✅ 34 untouched benchmarks

⚠️ Please fix the performance issues or acknowledge them on CodSpeed.

Benchmarks breakdown

	Benchmark	main	nameexhaustion:csv-file-list	Change
❌	test_groupby_h2oai_q5	7.7 ms	9.9 ms	-22.39%

[codecov]

Codecov Report

Attention: Patch coverage is 93.39623% with 14 lines in your changes are missing coverage. Please review.

Project coverage is 80.93%. Comparing base (81cc802) to head (dfc3bdb).
Report is 5 commits behind head on main.

Files	Patch %	Lines
...olars-lazy/src/physical_plan/executors/scan/csv.rs	92.37%	9 Missing ⚠️
crates/polars-pipe/src/executors/sources/csv.rs	92.53%	5 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main   #16180      +/-   ##
==========================================
- Coverage   81.03%   80.93%   -0.10%     
==========================================
  Files        1392     1393       +1     
  Lines      178939   179559     +620     
  Branches     2907     2907              
==========================================
+ Hits       144997   145323     +326     
- Misses      33436    33730     +294     
  Partials      506      506              

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

[nameexhaustion]

failing test is python import timing - I don't think this PR has anything to do with it

[ritchie46]

Can we be sure here the the DataFrames have the same Schema?

If so, we can use concat_df_unchecked. It isn't unsafe, but it will elide schema checks and duplicate name checks.

[nameexhaustion]

very good catch, I checked and realized this PR as it is would cause differing schemas to be silently coerced, I will need to do some more fixing

[ritchie46]

Looks great! Have a few minor comments.

I suspect this also should be much better in performance when dealing with many files.