Good day, Tung:
You are correct that none of the examples you shared is subject to the linked expression special margin rule of the Nemeth Code, because none of the examples you shared is displayed.
Yes, I would agree that you can use the "regular box" shape for the blue-shaded squares in your non-spatial problems and the general omission indicator for the blue-shaded squares in your spatial problems. This use should be explained in a transcriber's note.
In your message, you described the "regular box" shape as (dots 1246 246), but it should be (dots 1246, 256).
The last part of problem 4 in your examples is not spatial, as you said, and so technically the "regular box" shape could be used instead of the general omission indicator for the blue-shaded squares in this part. However, I would encourage you to use the same omission indicator throughout any one problem and to word your TN about what is used in braille to represent print's blue-shaded squares accordingly.
Thank you for the questions and for the examples!