Refactor the model chunking script to work with all version of coremltools

[jakesabathia2]

• In coremltools 8.0b2, they introduced an util ct.models.utils.bisect_model (Add bisect_model API under ct.models.utils coremltools#2286),
  in which does the same thing as chunk_mlprogram.py, with a more robust implementation.
  For instance, taking the source CoreML model spec version into account.
• This PR makes the chunk_mlprogram.py works on older coremltools, like 7.2, which has been reported a bug (chunk unet does not work with coremltool>7.1 #333).
• Also for the coremltools version >=8.0b2, we refactor the script to directly call into bisect_model API.
• The legancy implementation should be removed after we officially set the dependency on coremltools>=8.0

Testing:
python -m python_coreml_stable_diffusion.torch2coreml --convert-unet --convert-text-encoder --convert-vae-decoder --convert-safety-checker --model-version CompVis/stable-diffusion-v1-4 -o ../../model_zoo --chunk-unet
ran on both coremltools==8.0b2 and coremltools==7.2