Add folder-level zero-shot classification example

[arjxnt]

Summary

• Add a folder-level zero-shot classification example for MobileCLIP2/OpenCLIP inference.
• The script scores images in a directory against repeated --label values or a newline-delimited --labels-file.
• It writes top-k predictions and per-label scores to CSV, with batching and auto/cpu/cuda/mps device selection.
• Link the example from the README near the existing inference snippet.

Validation

• Ran: python -m py_compile examples/zero_shot_folder_classification.py