More details in their proposal here.

Can anybody say if this would then be better in any regards (throughout, latency, etc.) than Scylla (a C++ reimplementation of Cassandra)?

This doesn’t appear to be decided yet, but rather a tweet that links to a discussion.