Which strategy replaces instances with the new version in batches, needs little extra capacity, and has old and new versions serving traffic at the same time during the rollout?

A rolling deployment replaces a few instances at a time. So it needs little extra capacity, but both versions serve traffic during the rollout.

"Blue-green deployment" is wrong: it keeps a whole second environment and switches all traffic at once. "Recreate (stop all, then start all)" is wrong: it causes downtime, because nothing serves traffic between the stop and the start.