Currently, a subsystem only registers a single XMLElementWriter<SubsystemMarshallingContext>. Consequently, we must always register a writer for the minimum available stability.
e.g.
Subsystem "foo" defines 2 schema versions:

• wildfly:foo:1.0
• wildfly:foo:preview:1.0

When registering its writer, we register the writer based on Stability of the server and the available schema versions.
e.g.
Stability.DEFAULT -> registers wildfly:foo:1.0 writer
Stability.COMMUNITY -> registers wildfly:foo:1.0 writer (since there is no community-specific schema)
Stability.PREVIEW -> register wildfly:foo:preview:1.0 writer
Stability.EXPERIMENTAL -> registers wildfly:foo:preview:1.0 writer (since there is no experimental-specific schema)

Unfortunately, makes it problematic to when switching from one stability level to another.
e.g.
If I start the server using Stability.COMMUNITY, the subsystem schema will persist as wildfly:foo:1.0.
If I restart the server using Stability.PREVIEW, the subsystem will persist as wildfly:foo:preview:1.0.
However, if I have not changed this subsystem's management model, I cannot simply restart the server using Stability.COMMUNITY, since the previously persistent server configuration will fail to parse because the wildfly:foo:preview:1.0 is not enabled by Stability.COMMUNITY.

If we instead allowed subsystems to define a writer per Stability (like we already do for parsers), we could, prior to persisting a given subsystem, interrogate its management model and determine the maximal persistable stability of its defined resources/attributes. In our example, if the server is started using Stability.PREVIEW, but no preview resources and/or attributes are defined in its model, we can conclude that the foo subsystem is persistible using wildfly:foo:1.0.

Thoughts?