L2 cache: 512KB per core --> 256KB per core
L3 cache: 8 MB shared --> 4MB shared
And Voilá!, you have your zen-lite NEO edition!.
In fact, now the news has more sense than before. Zen is the most logical choice.
At least from the point of view of backwards-compatibility I think that Zen could be expected to run Jaguar code.
There might be some low-level differences that could lead to some optimizations like L2 cache blocking not scaling as expected, although that can depend on how fast Zen's last-level L3 is as the LLC versus Jaguar's L2.
There are some differences in core ID enumeration that might matter more for the OS, but if Zen's SMT capability were in play, attempts at pinning to specific cores might misbehave if the abstraction layer doesn't compensate.
What doesn't track with that currently is the rumor that there are no Neo-specific x86 paths, whereas DF reported that there would be separate graphics paths. There are instructions and features that Zen could offer over Jaguar that at least from that report would be left on the table.
Another item that might not track is what Zen-lite would be. Even if the cache were cut down, there's no rumor for or clear justification for AMD to have designed a significantly cut down server core that should be larger than Jaguar. I'm not sure there's much prospect for any die shrinkage if both the GPU and CPU sections expand so much.