Can an extended version contains 2-3 conference papers? If yes, should all authors be listed or only the key contributors?
An extended journal version might well combine multiple input papers, though this is somewhat unusual. In general if somebody was an author on one of the papers that is a source, then they should be an author on the composite. The only exception that I can think of is if the resulting composite paper drops sections from some of the inputs, and certain authors had only been involved in those sections. Again, this would seem to be a rather unusual situation, though plausible.