sam2 large video support

[hrnn]

• This is the initial effort to support large videos (longer than 1m).
• It splits the input video into segments, so instead of process the entire video, we process segments of it.
• The process_segment function handles the object detection within each segment, while the merge_segments function combines the detections from all segments. The merging process considers object overlap and identity across segments to ensure consistent tracking throughout the video.

Comparison between current result and the pull request result

final_output_pr.mp4

[camiloaz]

LGTM! I am assuming there were already tests for the object tracking case?

[camiloaz]

LGTM! I am assuming there were already tests for the object tracking case?

let's wait for @dillonalaird review.

[dillonalaird]

Added a few comments on how to handle the display data

[dillonalaird]

LGTM