Bug fixes + refactoring

[straeter]

• fixed a bug where skip_cleaning is applied at the wrong location (it skipped the whole data processing instead just the article cleaning)
• fixed a bug where the article cleaning dismissed the language information, so that only_en or only_zh had no effect
• fixed a bug where the default output file was f"{target_model}_raw_results.jsonl" but the model checked in "raw_results.jsonl" by default for existing data (now both contain the model name)
• added language key/value to raw_ and clean_ jsonl example data for claude-3-7-sonnet and to criteria.jsonl
• added a deep_research_bench_fact.py script that does all the utils steps + updated the bash script
• made result direction structure consistent between fact + race
• refactoring of the main script
• type hints + PEP8 style