Misalignment in index due to nissing cuda ctream parameter in vision embed_image

[fan-chao]

A bug was discovered during testing in the scan functionality. In nanodb.py at line 253, the vision model was not being passed a cuda stream parameter, which led to data misalignment during cuda synchronization. This resulted in incorrect final index data.

The corrected code is as follows:
embedding = self.model.vision.embed_image(data, stream=self.model.stream)

This modification ensures that the stream attribute of the model is correctly passed to the embed_image method, maintaining data integrity during the synchronization process and ensuring accurate index data.