add CLIPScore metric

[Sunhill666]

Fixes #705

This pull request adds a new CLIPScore metric for evaluating the alignment between images and their corresponding text descriptions. The implementation includes the metric logic, a Gradio-based web interface, and documentation. The most important changes are summarized below:

New Metric Implementation:

• Added clip_score.py implementing the CLIPScore metric using the CLIP model from HuggingFace Transformers, including logic for computing the similarity between image and text pairs.

Web Interface:

• Introduced a Gradio app in app.py that allows users to interactively compute CLIPScore by uploading an image and entering text, with example inputs and integration of the metric.

Documentation:

• Added a comprehensive README.md describing the metric, usage instructions, input/output formats, example code, and citation information.

Dependency Management:

• Updated requirements.txt and setup.py to include the necessary transformers library for CLIPScore functionality.