Refine Your Accepted Paper: Maximizing Changes Before Camera Ready

We have a paper accepted to NeurIPS, but at the same time we were working on a resubmission to ICLR just in case NeurIPS rejected us. There has been substantial rewriting, and we feel it would be a waste if we discarded all of it. To give a summary of what's changed:

• We completely rewrote every single section except for the results and conclusion. We even changed the paper structure.
• Intro, related work, and background knowledge were completely rewritten to avoid confusion.
• Method now has a pipeline graph, and all the text detailing each block in the graph. Previously, it was dumping formulas, so the entire section has been rewritten.
• We also added some scaling and smoothing to our algorithm so our method is more stable. But this changed a lot of our hyperparameters and the sensitivity study's graph. (The entire shape of the graph changed)
• We added 1 new theorem with 5-page proofs in the appendix. This came from one of the attacks by a reviewer, we answered the attacks by proposing 1 new proposition during the rebuttal. But when we formally wrote it down, it turned into a full theorem with a 5-page proof. This would have changed our entire theoretical contribution. We really don't want to discard it, but not sure if we can add something this big in the camera-ready.
• Remove 1 word from the title. Change our theoretical contribution, but the method and empirical contribution remain the same.
• Added about another 5 extra pages in the appendix explaining experiments and metrics (reviewers asked for them). So 10 extra pages in total.

Does anyone know how much change for camera-ready is acceptable? Can a paper get rejected if we change too much during camera ready or they will just tell us this is not acceptable please re-submit something closer to the version during review?