I faced this exact issue on a compact optical payload project last year. We ended up using a nested gimbal ring setup with slim-section crossed roller bearings and direct-drive motors. The inner ring handles pitch, while the outer ring handles yaw. Since both rings are supported by precision bearings, the backlash is virtually eliminated and the whole mechanism fits within a 58 mm cube.

I would avoid using plastic 3D prints for the rotating elements. We machined everything from aluminum and used hard stops for safety.