Warning! Python 3.9 is not supported by Tensorflow officially. Some calls may fail. The use case of these packages is to provide a developing environment(E.g. Jedi support) for those who use python 3.9 as the python interpreter.
However, it seems ok to use these packages in a development environment where only auto-completion and lint are needed. Tested under vxst-workspace docker image, and it works.