Hmm. There's a multitude of questions and issues being tossed out here. But, albeit not documented in a great way in their online manual and faq etc., DSS does exactly what it is supposed to, if you understand it and load your files properly. I readily admit to doing it wrong early on when I first tried mutli-session imaging. Also, you must make sure that what you are intending to do actually makes sense. Otherwise you can just scramble your master files and end up with poor output.
First, you pretty much do not want to feed a finished stack back into DSS as a sub, to be combined with new individual subs. There's a good chance it won't even work, especially if the image size no longer matches (from intersection mode, for example). You also may have applied various settings in stacking that have altered the image significantly from what the subs are. Finally, while you can do "stacks of stacks," I think it is generally considered best practice to load all individual subs in again as groups, in order to reap the benefits of your rejection algorithm. The one caveat here is that you might be able to reuse masters. That is not guaranteed, as there are changes that can result in new masters being needed. For example, if you have a saved list and register it using bias frames, the other master calibration frames will incorporate that bias (or master bias), and overwrite them if masters already exist. Those masters can no longer be used in a stacking scheme that does not include that bias (for example if you wanted to try it with dark flats in lieu of bias).
Finally, multi-session group stacking works fine, and all masters for each session are created appropriately. Just remember that, for group stacking, any calibration frames in the main group will be applied globally. And as noted above, if you set it up wrong, the application of those files can mangle your other masters. In the usual scheme of things, which is same target over different nights, the only file you will have in Main is the bias or master bias. And if you don't use bias, Main will probably be empty. Then Group 1, 2, etc. will each have its own set of lights and associated calibration frames. Also, it will behoove you to be organized with each night having its own folder, with subfolders within for lights and the different calibration subs.
You can even do a quick test of this if you set it up in some new folders, say 5 subs of each, and watch it work on the little window that pops up. During registration it will go through and create masters for each group.
I believe it is still generally best to stack same exposures and iso even using groups. It will allow you to do otherwise, but you have to make sure it is appropriately weighted if need be. Or, you might be able to use one of the background calibration schemes so that the differing groups are stackable without the result being junk.
DSS does have an "entropy" stacking mode that is specifically made for combining different exposures and iso's. I haven't tried it, so can't add anything there. It may be something to try if that kind of "HDR" combination of exposures is what you are after.
Hope this helps.