- Remove references to `SQLContext` as that was deprecated in Spark 3.0
- Change `lr` to `learning_rate` in all `SGD` arguments
- Pin the Tensorflow version to <=2.10, as 2.11 results in `weakref` errors in serialization (https://github.com/danielenricocahall/elephas/issues/16)
- Enhance typehints and docstrings