It's definitely important to discuss. It's a bit complex.
Any spec project that removes the SM calls from their API jars is making a breaking change, so they need to update their version accordingly. I believe by releasing a new major version. If an application that called into that spec API could run with the SM enabled, and then they update the spec API artifact and they get SM failures, that is a breaking change.
AFAIK, there are no plans to break the ability to call the SM APIs in SE 25, beyond a change to java.security.Policy.setPolicy where it now fails if invoked. Other than that JEP 486 doesn't remove the SM-related APIs typically used by libraries or cause them to fail. They just no longer do any enforcement. Since the user can no longer even enable the SM, there is no enforcement to be done. The upshot of this is a library that wanted from the same binary to support both SE 25 with no SM support and SE 21 with SM support could do so. Removing the internal SM calls means they need to maintain a new branch.
However, it is clear Java SE wants the broader Java ecosystem to move on from the SM, and personally my bet (based on ZERO information) is that the APIs will be gone by SE 29. So specs are well served by getting on with removing their SM code.
Further, I think it's a great thing for EE how recently it has been possible to certify an EE impl on an SE version that was not around when the EE version came out. WildFly was able to certify as EE 10 compatible on SE 21. I think it would be great if EE 12 impls could certify on SE 29. But if the SM API are removed in SE 29 and a spec API hasn't removed their calls, that is not possible. At least not without someone forking the spec API and producing their own artifacdt.
Then there are other situations like we saw with Connector where the spec itself has language that assumes SM support. That clearly needs updating.