[BUG] Convert OpenCLIP embeddings to numpy arrays (#3051)

## Description of changes
OpenCLIP embeddings are failing type validation, as they return a
`Tensor`. Converting to NuPy arrays to fix.

## Test plan
- [ ] Tests pass locally with `pytest` for python, `yarn test` for js,
`cargo test` for rust

## Documentation Changes
To discuss

chromadb/utils/embedding_functions/open_clip_embedding_function.py

@@ -59,13 +59,13 @@ def _encode_image(self, image: Image) -> Embedding:
                 self._preprocess(pil_image).unsqueeze(0)
             )
             image_features /= image_features.norm(dim=-1, keepdim=True)
-            return cast(Embedding, image_features.squeeze())
+            return cast(Embedding, image_features.squeeze().cpu().numpy())
 
     def _encode_text(self, text: Document) -> Embedding:
         with self._torch.no_grad():
             text_features = self._model.encode_text(self._tokenizer(text))
             text_features /= text_features.norm(dim=-1, keepdim=True)
-            return cast(Embedding, text_features.squeeze())
+            return cast(Embedding, text_features.squeeze().cpu().numpy())
 
     def __call__(self, input: Union[Documents, Images]) -> Embeddings:
         embeddings: Embeddings = []