EasyDel implementation of Clip JAX model

[ctr-pmuruganTT]

Problem description

Modify loader of CLIP Jax model to use EasyDel Instead of Transformers library

What's changed

• Modified the model source
• Removed load_processor and used AutoImageProcessor
• Added get_input_activations_partition_spec and load_parameters_partition_spec

Checklist

• New/Existing tests provide coverage for changes

[vzeljkovicTT]

Using CLIPModel is appropriate when we already have EasyDeL formatted weights. In our case, we are loading weights directly from HuggingFace and converting them into JAX NumPy arrays.
Easydel provides AutoModel family of classes which is exactly for this purposes (model params taking from HF and transfer to jax arrays; after that calling model implementation e.g. ClipModel).
For this use case we should use AutoEasyDeLModelForZeroShotImageClassification (from easydel import AutoEasyDeLModelForZeroShotImageClassification).

[vzeljkovicTT]

Why did we deleted this _load_processor? It was better to use AutoImageProcessor instead of CLIPProcessor?

[vzeljkovicTT]

This flag from_pt is unnecessary in the easydel