Add sv.Classifications.from_clip data loader

[capjamesg]

This PR adds a new annotator for annotating classification results. This PR also adds a new data loader that enables you to load CLIP classifications into a sv.Classifications object.

Testing code

import supervision as sv
import clip
from PIL import Image
import torch
import cv2

SOURCE_IMAGE_PATH = "cafe.jpg"

image = Image.open(SOURCE_IMAGE_PATH)
model, preprocess = clip.load('ViT-B/16')

image = preprocess(image).unsqueeze(0).to("cpu")

text = clip.tokenize(["a piano", "a cafe"])

with torch.no_grad():
    logits_per_image, logits_per_text = model(image, text)

    classifications = sv.Classifications.from_clip(logits_per_image)

    # should show two results, indicating all predictions are loaded into the object
    print(classifications.get_top_k(3))

annotator = sv.ClassificationAnnotator()

image = annotator.annotate(
    scene=cv2.imread(SOURCE_IMAGE_PATH),
    classifications=classifications,
    labels=["a piano", "a cafe"]
)

sv.plot_image(image)

Here is an example of the annotator in use:

Here is the raw image on which inference was run with CLIP in the code above:

[capjamesg]

Colab for this PR: https://colab.research.google.com/drive/1yC4pQ7AAtKrwQZ61XszACWKzNo8sNgwk#scrollTo=xlPWe7fZezr0

[SkalskiP]

Awesome @capjamesg 🔥 Merging!