Add support for Yggdrasil Decision Forests models

[achoum]

• Add support for Yggdrasil Decision Forests models.
• Remove support for TensorFlow Decision Forests models. Note: TensorFlow Decision Forests is powered by Yggdrasil Decision Forests models, and TensorFlow Decision Forests models are Yggdrasil Decision Forests models.
• Add a "Development" section in README.
• Update test imports so "pytest" can be run from the project root directory.

[thesps]

Thanks a lot for this contribution, and sorry for the delay in reviewing it. To me the code is fine as is, I don't propose any changes. I've checked the tests and examples and I think all is looking good. Thanks also for the general improvements, I appreciate it.

Merging!

One query I have, that we can come back to later, is: how difficult would it be to extend support to other tasks? (if model.task() != ydf.Task.CLASSIFICATION here)

[achoum]

Thanks for the merge :).

how difficult would it be to extend support to other tasks?

It would be trivial. This current code would likely immediately work for regression and ranking (to be tested).

Background

In the case of GBDT, trees have the same structure for all the tasks. What changes is how the sum of the tree predictions is used (i.e., the link-function, a.k.a. activation function). Since conifer does not apply a link function, this code is more-or-less task independent.