Add weight streaming at runtime

[eddieliao]

Figure out a way to have weight streaming at runtime i.e. be able to fit large models on gpu without needing to know literal size ahead of time

• [x] Define/determine an allocation of literals to be streamed
• [x] Move copy instructions to separate stream
• [x] Investigate why @literal instructions take up so much time
• [x] Decrease time spent on @literal instruction
• [ ] Move from naive allocation to "smart" allocation (figure out best way to mask time taken to copy)