Supose you have to clients who send each a separate SRT stream to SRS, without streamid info - ex. OBS/ffmpeg srt://srs-server-ip:port?mode=caller. Either they don't know how to use streamid or they use hardware encoders which does not have streamid implemented (firmware with older SRT lib version).
So how clients can request in that case one or the other of the two streams?
Both are ending in SRS with the default "#!::r=live/livestream" streamid! I doubt that SRS will even properly deal with data from both streams simultaneous.
If SRS is receiving already a SRT stream without streamid info the second one will be ignored ( connection won't even be made/accepted for the second stream). ;)
I think that we need a proper method to identify and accept srt streams, even without streamid info.
regards,
Laur

What happens if the client's IP changes or we don't know the IP when the client is behind NAT? What should we do if we need to add a new client—should we stop existing streams?

I think the use case makes sense, but we shouldn't assume we use the client's IP to generate the streamid. A better solution is to use HTTP hooks. If there's an empty stream ID for SRT, let your service generate the stream ID based on the client's IP or other information.