Easier support for local training

[SteffenRenisch]

Application code in standalone mode may use python packages from different installation than when executed within NVFlare.
Also the entry point is different in standalone mode vs. when executed from within NVFlare.
Do newer version of NVFlare provide better support for it.

[oleschwen]

Please clarify.
The docker.sh scripts in the startup kits (intended starting point) can run a local training or start a swarm node where the "same" code is executed in a Docker container using the same image with the same python packages.
One can start the container in interactive mode and deliberately do something different. One can execute the application code outside Docker in arbitrary Python environments. I doubt it would be feasible or desirable to prevent these possibilities.
We could possibly simplify installing the same environment locally (outside Docker), but should clarify what to achieve.