For the second part, using separate audio tracks in OBS, I’ve had a good time just using pactl commands to create the virtual devices, route it in qpwgraph and then save that as a preset. When qpwgraph launches at startup it loads that routing how I left it.
For discord screensharing there’s alternatives like discord-screenaudio and webcord (which has a flag you can use to enable audio sharing). Both of those are able to send audio to the proper stream audio through discord so your friends don’t hear it through your mic, but they have to be based on the web version of discord so there’s no hardware encoding. If your plan is to stream games and stuff, fast motion will cause the stream to be very choppy for viewers. So in that case using the native discord app and routing audio through your mic source is kinda the only way to do it as of now, unfortunately.