I don't know about roslaunch's internal implementation, but I think most of these should be easy to implement as services (maybe changing output to screen during runtime is not possible).
Use cases for this come up when running a large roslaunch and something unexpected happens. If that node is somewhere in a roslaunch, the only chance to interact with that is by whatever the node provides currently. Especially if respawn is true, it is not even possible to manually kill the node and start a new one indenpendently. Although that would obvisouly be possible by killing the complete roslaunch, restarting that might take a lot of time and the situation that caused the problem might not exist any more.
It also would enable nice GUI interfaces. node_manager_fkie seems to have some features, but I am not aware how/if they interact with roslaunch.