Add support for py312, drop support for py38

[wbenoit26]

This PR expands the python version range to include 3.12. As a consequence, the TensorFlow version range has to be changed to include 2.16, as it's the first version that includes support for python 3.12. However, TensorFlow 2.16 is also the first version that uses Keras 3 by default. I'm not sure how feasible it is to support both Keras 2 and 3, or whether it would be worth the effort of doing so, but it's possible to set an environment variable that forces the use of Keras 2.

Edit: And to use the tf-keras package that allows the continued use of Keras 2, python 3.8 can no longer be supported.

[wbenoit26]

@EthanMarx This should be good to go, let me know what you think.

[EthanMarx]

What is the reason we need to add tf-keras as a dep? can we get away with not specifying it and letting users decide which version they use? I'm not as familiar with Keras

[wbenoit26]

Using Keras 3 broke some of the tests - it seems to have changed the way layer names get used, and I'm not sure there's a syntax that works for both Keras 2 and 3, though worth looking into in the future.

The tf_keras package is just Keras 2.15 with a different package name (see point 1 under "Moving from Keras 2 to Keras 3" here), and it's how Tensorflow recommends using Keras 2 with Tensorflow 2.16.

[EthanMarx]

Okay got it - so migrating to Keras 3, will break the export. Definitely makes sense to hard enforce keras 2 then.