Add `CLIP` model

[james77777778]

Related to #1752

Colab for demonstrating the prediction of the ported backbone: https://colab.research.google.com/drive/1MgrQ1jq8wcICfoSbxp075wfap2qYADGs?usp=sharing

Preset: openai/clip-vit-base-patch32 (should work for all CLIP models)

Inputs:	Image	Text
	"a photo of a cat", "a photo of a dog"

Outputs (probability):

• huggingface CLIPModel: 99.5% vs. 0.5%
• KerasHub CLIPBackbone: 99.7% vs. 0.3%

There are some questions about the upcoming task definition:

• What type of Task should we introduce? The original FeatureExtractor seems somewhat ambiguous to me.
• Should we modify CLIPPreprocessor to accept image-text pairs for the new CLIPBackbone or new task? If so, we would need to update the SD3 implementation as well.

cc @divyashreepathihalli

[james77777778]

Can you please add the presets and the conversion script? It could be in a follow up PR.

Sure. Will submit a PR soon. However, the questions remain:

• What type of Task should we introduce? The original FeatureExtractor seems somewhat ambiguous to me.
• Should we modify CLIPPreprocessor to accept image-text pairs for the new CLIPBackbone or new task? If so, we would need to update the SD3 implementation as well.