Decoupling tensorflow from keras in stubs

Whilst working on https://github.com/python/typeshed/pull/11696, I was having issues referencing the source code. After reading the doc and a bit of investigation, I realized why:

The runtime of https://github.com/python/typeshed/tree/main/stubs/tensorflow/tensorflow/keras actually are just references to the keras source code at https://github.com/keras-team/keras/tree/master/keras (import keras), and is generated / injected in import tensorflow.keras by https://github.com/keras-team/keras/blob/6454a4888a494c20ab0ea1dc6912cbcc13c5f940/pip_build.py#L62

Even tensorflow's own doc links back to karas source code https://www.tensorflow.org/api_docs/python/tf/keras/callbacks/Callback

>>> from keras import callbacks
2024-04-03 23:28:35.387416: I tensorflow/core/util/port.cc:113] oneDNN custom operations are on. You may see slightly different numerical results due to floating-point round-off errors from different computation orders. To turn them off, set the environment variable `TF_ENABLE_ONEDNN_OPTS=0`.
2024-04-03 23:28:36.959410: I tensorflow/core/util/port.cc:113] oneDNN custom operations are on. You may see slightly different numerical results due to floating-point round-off errors from different computation orders. To turn them off, set the environment variable `TF_ENABLE_ONEDNN_OPTS=0`.
>>> callbacks.__file__
'...\\typeshed\\.venv\\lib\\site-packages\\keras\\callbacks\\__init__.py'
>>> from tensorflow.keras import callbacks 
>>> callbacks.__file__
'...\\typeshed\\.venv\\lib\\site-packages\\keras\\callbacks\\__init__.py'

In other words, keras "pollutes" the tensorflow namespace! I don't know if the type stubs specifications allow two distributions to write to the same stub-only package, which would allow us to perfectly reflect what's truly happening. Second best thing would be to have stubs for keras and keep tensorflow.keras in the tensorflow stubs (since tensorflow is dependent on keras anyway)

CC @hoel-bagard & @hmc-cs-mdrissi

Lisez stubs/tensorflow/tensorflow/keras ainsi que le fichier pip_build.py de Keras et les sources d’exécution liées pour comprendre comment tensorflow.keras est injecté. Confirmez auprès des maintainers l’organisation prise en charge des stub packages ; la tâche est terminée lorsque la disposition choisie représente correctement à la fois keras et tensorflow.keras sans ambiguïté de namespace.