Upgrade MongoDB Atlas vector search integration to use vectorSearch type

[brimoor]

Change log

• upgrades the MongoDB Atlas vector search integration to use the generally available vectorSearch type rather than the beta knnVector type
• fix bug with similarity queries on grouped datasets

Example usage

export FIFTYONE_DATABASE_NAME=fiftyone
export FIFTYONE_DATABASE_URI='mongodb+srv://$USERNAME:$PASSWORD@$NAME.mongodb.net/?retryWrites=true&w=majority'

import fiftyone as fo
import fiftyone.zoo as foz
import fiftyone.brain as fob

dataset = foz.load_zoo_dataset("quickstart-groups")

view = dataset.select_group_slices("left")
index = fob.compute_similarity(
    view,
    model="clip-vit-base32-torch",
    brain_key="img_sim",
    backend="mongodb",
    embeddings="embeddings",
)

assert index.ready

print(index.total_index_size)  # 200

view = dataset.sort_by_similarity(dataset.first().id, k=5)

view = dataset.sort_by_similarity("kites high in the sky", k=5)

[allenleetc]

Code LGTM

Tested with grouped image dataset. Search is fixed when searching the original grouped dataset, both in-App or via SDK. Searching a flattened view (eg generated via dataset.select_group_slices() ) fails with a different, probably unrelated, error.