model: Video support

[johnandersen777]

This is a meta issue to track figuring out how we are going to do a clean implementation of support for videos.

Please comment any thoughts / links to branches or anything you've thought and played around with related to this!

[johnandersen777]

We've talked about two approaches so far:

• Records a whole video
  • We probably need to add an async methods to record to fetch the next frame in this case. In case video is live streamed.
• Records as frames
  • This might require less / maybe no modifications to the Record class. However, it introduces the problem of how does a model know when a frame belongs to one video vs. the next video when training.

[mhash1m]

Perhaps we could make use of directories for every video or just naming through identifiers for the frames eg. vid1frame1, vid2frame1 to differentiate.