Hidden requirement of tensorflow version should use extras_require

I'm noticing issues very time we update the requirement of tensoflow for this check:

https://github.com/tensorflow/probability/blob/40b18076b53d7c6e22e1fa5f95d805be0cfa0cc1/tensorflow_probability/python/__init__.py#L52-L64

It would be nice if we could add an option to extras_require so that it was more like:

extras_require={  # e.g. `pip install tfp-nightly[jax]` 
     'jax': ['jax', 'jaxlib'], 
     'tfds': [TFDS_PACKAGE], 
     'tf': ['tensorflow>=2.8'],          # <--- Add a tf option here 
     'tf_gpu': ['tensorflow-gpu>=2.8'],  # <--- Add a tf_gpu option here 
 } 

This way if somebody is using both tensorflow and tensorflow_privacy, they can put:

# requirements.txt
tensorflow >= 2.6
tensorflow_privacy[tf]

and there will be no package conflicts after install.

Relevant: https://github.com/tensorflow/tensorflow/issues/7166

Start with the dependency-version check in tensorflow_probability/python/init.py and inspect the package configuration that defines extras_require. Compare the requested tf, tf_gpu, jax, and tfds extras with the existing dependency setup; done means the TensorFlow extras express the supported requirements without the reported package conflict.