IMHO Cartesi was always about the execution layer, after all the core tech is a VM. But remember that not so long ago blockchain modularity and “layers” were not a thing.
The goal was to provide computationally scalability, so the project focused on that. And I still think Cartesi still have the best VM out there that has the required properties to power blockchain architectures.
But it was also predictable that computation scalability brings along necessity of data availability scalability (check the cone of innovation diagram).
But DA was never Cartesi focus. Noether was conceptualized, even some code was produced, but other projects took over that goal, which is not a simple problem at all.
I don’t think forking an existing project (e.g. Celestia) is the best alternative. Forking requires huge efforts. I think the best alternative is integrating, even with the risks you describe. Some experiments and discussions are already in progress.