Removing Tensorflow from code base

[JaMe76]

In recent months, several libraries that deepdoctection depends on have removed models previously implemented in TensorFlow. For Document Layout Analysis, deepdoctection relies on the Tensorpack library (for model construction and training), which was originally designed for TensorFlow v1.

With the latest deepdoctection release (v0.45), it has become clear that installing a suitable TensorFlow version and aligning it with the necessary dependencies (e.g., numpy v1) is increasingly diverging from cutting-edge versions.

This poll is intended to gather feedback on whether users still want the library to continue relying on TensorFlow. The integration of new models and developments in TensorFlow has not been pursued in deepdoctection for several years.

The purpose of this poll is therefore to determine whether support for TensorFlow models should be dropped entirely by the end of this year.