fix(ui): render structured benchmark details correctly

UI polish: table options grouping, column visibility redesign, model size fixes

added filters and fixed adaptive filtered average

fix to show bert score for each sample

Align FannOrFlop detail summary with results f1 and return full benchmark-detail rows

fix issue with model's type emoji

Add Expand and Collapse button for long samples in the detail section

Enahnce downloading the details repo

Fix score mismatch between details and table

Prioritize extracted_json over raw_response

Enhance Details for benchmarks without sub-tasks

New formats for details, add fannorflop benchmark

fix issue with multi-options answers, and with the samples that don't have binary score

fix details

fix issue with some parquet files have different formats

fix requirements

adding details

fix get_model_size

change the way we read results by specifying source type, source type is the prefix of the filename. Sperating the tasks into different lists based on the source

fix some of the status

fix requirements.txt

feat: add average column and improve file loading logic

Add ArabLegalQA and MedArabicQA benchmarks + default missing results to 0

download_datasets

push first code