The code from orchardantispam will be used, because what is does is to provide compatibility with other providers, and also because it is not tight to Comments.
As Bertrand is suggesting, there should be some events to trigger spam check. So if no spam module is installed, it wouldn't interfere with the caller module. Also multiple spam modules can act during the process. If one fails then it's spam. Those modules
could also be extended to provide statistics or content retention. It should also handle specific rules like white/blacklisting specific users/domains/
I agree that maybe the comment module should then be migrated to use the new infrastructure.