Added wds/voc2007_multilabel

[djghosh13]

Finishing off support for all datasets currently supported through the original interface:

• Multilabel datasets can be converted to and loaded from webdataset
• Converted and uploaded voc2007_multilabel to HF
• Accuracies match within 0.0001

You can test the new dataset with: (replacing model with whatever you like)

clip_benchmark eval --dataset wds/voc2007_multilabel --dataset_root "https://huggingface.co/datasets/clip-benchmark/wds_voc2007_multilabel/tree/main" --model "ViT-B-32" --pretrained "openai" --output results_voc2007_multilabel.json