why would you store a spammy note? or you mean maybe like a bayesian spam probability value... still even there, you'll set a threshold and drop those ones, and clients aren't going to implement such spam filter confidence value settings any time soon unless you build/fork one that does

the rest can be filtered out using a regular filter, though you need to create a second mode for filter matching that is a blacklist, the standard filter mechanism is a whitelist match...

this is something that probably even needs to be properly formalised, i saw someone made a NIP proposal, maybe it's NIP-66, that uses a ! prefix to negate matches, could be extended to filters to have negation in the filter fields - though again... needs to be in the client side also, although there is some reason to apply these based on privilege or other things that may relate to a user account (eg, don't ever send NSFW tagged option in a relay subscription config)