I fully support this proposal, this is the way to go!
It should be noted that there is a way of using sequencers right now with the current Cartesi Rollups release: just deploy as an L3 on top of an L2 that does have a sequencer (like Optimism or Arbitrum). While this does give DApps lower latency and and costs, it has some very important drawbacks, like depending on a specific L2 and not being able to jump into the shared sequencer bandwagon - which aside from promising better L2 <> L2 integration, often comes bundled with extended DA capabilities (a must IMO)
Implementation-wise, IMO proper sequencer support should be done by allowing DApps to access base layer state via the Dehashing Device, as described here.