Add option to save cached embeddings to disk for re-use

[thompsonmj]

To follow-up #16 and the discussion in bioclip, create an option for the text embeddings cache to be saved to disk and an option to load these from a specified filepath rather than recomputing if the category list is reused.

[johnbradley]

This can be somewhat done with the current code, but it isn't very appealing. The CustomLabelsClassifier class has two properties that contain the embeddings. classes - a list of class names and txt_features. Both of these items would need to be saved and restored.

Rough Code

Embed text an save "classes.json" and "txt_features.npy":

from bioclip import CustomLabelsClassifier
import numpy as np
import json
classifier = CustomLabelsClassifier(cls_ary=["dog","cat","fish"])
with open("classes.json", "w") as outfile:
    json.dump(classifier.classes, outfile)
np.save("txt_features.npy", classifier.txt_features.numpy())

Load "classes.json" and "txt_features.npy" and make a prediction:

from bioclip import CustomLabelsClassifier
import numpy as np
import json
classifier = CustomLabelsClassifier(cls_ary=[""])
with open("classes.json", "r") as infile:
    classifier.classes = json.load(infile)
classifier.txt_features = np.load("txt_features.npy")

print(classifier.predict("Ursus-arctos.jpeg"))