feat: Raise more informative error message for directories containing files with mixed extensions

[nameexhaustion]

If the user passes a single directory to a scan_* function, we will now check that all files underneath it have the same file extension. If this is not the case an error message is raised showing the offending paths and recommending to use glob patterns if they still wish to scan all files.

Combining this with #17478 will also make it so that file extensions of empty files are ignored.

Fixes #17436

[codecov]

Codecov Report

Attention: Patch coverage is 97.43590% with 1 line in your changes missing coverage. Please review.

Project coverage is 80.47%. Comparing base (553f30b) to head (64c38a2).

Files	Patch %	Lines
crates/polars-lazy/src/scan/file_list_reader.rs	97.05%	1 Missing ⚠️

Additional details and impacted files

@@           Coverage Diff           @@
##             main   #17480   +/-   ##
=======================================
  Coverage   80.46%   80.47%           
=======================================
  Files        1483     1483           
  Lines      195138   195159   +21     
  Branches     2782     2782           
=======================================
+ Hits       157014   157047   +33     
+ Misses      37612    37600   -12     
  Partials      512      512           

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

[alexander-beedie]

If you call scan_parquet, wouldn't it be cleaner to just load all the parquet files found in the given directory? It's pretty common for there to be an additional file representing a checksum, metadata, etc, alongside the parquet payload, and the caller's intention seems clear (given that they called a function that can only load parquet - or the filetype of the given scan_* function).

[ritchie46]

We decided not to determine what parquet files are by checking whitelisted extensions. E.g. parquet, par, pqt.

I also don't want to check for magic bytes on all the files in a directory, as that would require downloading all of them potentially, and this is impossible for (compressed), csv, json.

The idea is that if you pass a directory, you guarantee it is a (hive partitioned) dataset. If you want to load all files with a certain file extension, we give you the possibility to do so via globbing patterns.