Fixing custom embedded documents regression

[brimoor]

Fixes a regression introduced in v0.16.6 when working with custom embedded documents that would raise an error when a custom EmbeddedDocument class was used to populate a dynamic attribute of a Label field:

from datetime import datetime

import fiftyone as fo
import fiftyone.core.fields as fof
import fiftyone.core.odm as foo

class CameraInfo(foo.EmbeddedDocument):
    camera_id = fof.StringField(required=True)
    quality = fof.FloatField()
    description = fof.StringField()

class LabelMetadata(foo.DynamicEmbeddedDocument):
    created_at = fof.DateTimeField(default=datetime.utcnow)
    model_name = fof.StringField()

sample = fo.Sample(
    filepath="/path/to/image.png",
    camera_info=CameraInfo(
        camera_id="123456789",
        quality=99.0,
    ),
    weather=fo.Classification(
        label="sunny",
        confidence=0.95,
        metadata=LabelMetadata(
            model_name="resnet50",
            description="A dynamic field",
        )
    ),
)

dataset = fo.Dataset()

# Would raise an error on v0.16.6; now works
dataset.add_sample(sample)

[benjaminpkane]

LGTM