feat: add ImgEdit benchmark with edit type subsets

[davidberenstein1957]

Summary

Add ImgEdit benchmark for image editing evaluation with 8 edit type subsets.

Dataset

• Source: sysuyy/ImgEdit (HuggingFace) + GitHub JSON from PKU-YuanGroup/ImgEdit
• Edit types: replace, add, remove, adjust, extract, style, background, compose
• Collate: prompt_with_auxiliaries_collate

Implementation

• Add setup_imgedit_dataset in src/pruna/data/datasets/prompt.py
• Support subset param for filtering edit types
• Register in base_datasets
• Add BenchmarkInfo entry with metrics: ["accuracy"], subsets list
• Auxiliaries should include image (input), judge_prompt, subset
• Add test

Acceptance

• PrunaDataModule.from_string("ImgEdit") works (all subsets)
• PrunaDataModule.from_string("ImgEdit", subset="replace") works
• Auxiliaries include image, judge_prompt, subset fields
• Test passes