Update code for latest occwl version

[JoeLambourne]

Why?

In #11 and #12 we see that updates to the occwl library introduced breaking changes which were not spotted in the example notebooks. Thank you to @cteqeu for reporting the issues and @akashaero for sharing the workaround.

This PR addresses the problems

What?

• A new version of occwl is released see here. We now fix the version of occwl to this version to avoid further unexpected breakage.
• In jupyter_segmentation_viewer.py we remove the calls to the removed functions.
• All the notebooks are rerun and retested.
• Some warnings were firing in pytorch-lightning. To ensure that all metrics are computed correctly we pass the number of faces to the logger as the batch size. Notice that the IoU and accuracy metrics reported in the paper are per-face rather than per solid.
• In scale_utils.py we make use of the scaling function in occwl rather than re-implementing it here.