1) is difficult to answer, it depends what you want to do with the mask. If you've asked about the mask for model specification, this is typically not required / automatically done. Go with "Threshold masking" set to "Absolute" and then enter a threshold value of 0.1 or 0.2. This means the model will be based on those voxels with have a value of 0.1 or 0.2 at least in all of the subjects (which would correspond to a probablity of 10 or 20% for GM in your case probably).
Concerning 2) , you might want to use the usual templates or create a study-specific template averaged across the different normalised anatomical volumes (make sure they were normalised with the parameters/flow fields of the segmentation step you've already conducted). Instead of the anatomical volumes you could also use the unmodulated or modulated grey matter files of course, it might also be useful to rely on the average of the smoothed versions (as forwarded into the analysis). A study-specific template has some advantages I'd say, as the reader gets an impression of the quality of the segmentation.
Hope this helps,