Hi Anders!
1) https://picard.musicbrainz.org/ and https://musicbrainz.org/ are a database that also supports acoustic fingerprints (I imagine they have easy tools to create them). This is pure speculation on my part, but it sounds like the right thing.
2) All file typically used for music distribution (mp3, opus, flac etc.) have embedded meta data. Of course you know the title, artist and album. But there is more, maybe not directly a license field but at least some free-form commentary.
I would put the CC attributions in these metadata fields. Of course, they can get lost by format conversion or somebody deleting them before redistribution, but at some point you have done enough to satisfy the intent of the law and license.
3a) If you worry about people taking your music, you could audiomark it, like a secret fingerprint in the music. I use this: https://github.com/swesterfeld/audiowmark . The information is very robust inside the audio waveform, so people can'T delete it. They won't even know it is there. This is a very robust process that survives cutting, remixing, compression and format conversion. I have written down some experiment with it in the past, but it is in German. Maybe auto-translate can help:
https://osamc.de/archiv/audiowmark/
3b) Worrying about AI stealing your music... I don't have a solution for that; I don't think there is any. AI learning is a highly transformative process akin to humans listening to your music and learning from it.
I hope this brings you one step closer to your goals!
Nils