generate_image_prompts.py unnecessary txt embeddings

[genislm]

Descriptilon:

Deleted unnecessary ' ' (text) embeddings that cause a dimension error when stacked to image embeddings.

Explanation

The script to generate image embeddings was reclycled from the one used for text embeddings generation.
This causes not only unnecessary text embeddings but a bug in which a dimension error shows up.

To be more specific, the line 37 that said:
txt_feats = txt_feats.reshape(-1, txt_feats.shape[-1]).cpu().data.numpy()
Has the correct dimensions to be stored directly. However, after appending it to a list and np.stack it with the images embeddings (notice the [0] before converting it to numpy)
img_feats = img_feats.reshape(-1, img_feats.shape[-1])[0].cpu().data.numpy()
the dimensions of the np arrays to stack were not compatible.

This can be solved in two ways.

1. change the text embedding for
   txt_feats = txt_feats.reshape(-1, txt_feats.shape[-1])[0].cpu().data.numpy() (notice the [0])
2. Directly avoiding the text embedding, as I suggest.