Store transformed values in duckdb index file (#2737)
* index string lengths
* add image widths&lengths + refactor a bit
* add check for actul values in result (for string type)
* fix lists + keep None values in audio and image
* fix processing None values in audio and image
* treat cases when all values in column are None
* query specific columns in search (instead of data.*)
* get unsupported columns once
* rename columns __hf_length -> .length, __hf_duration -> .duration and so on
* refactor: move stats utils back to worker but outside of job runners
---------
Co-authored-by: Sylvain Lesage <sylvain.lesage@huggingface.co>
