1) Yes your observations are correct.
2) The standard global XYZ coordinate system makes sense when you think in terms of a first person shooter, when you are looking through the eyes of a character in the scene with a blank(identity) transformation matrix. Like it would when you draw a coordinate system on a piece of paper, X points to right and Y points upwards. According to the right hand rule (x->thumb, y->index finger, z->middle finger), Z points towards you.
3) It wouldn't be wrong, but it would be a diversion from standards. There are three problems that I can think of at the moment: (a) Let's say one day you want to use a physics library that uses the standard coordinate frame. If you did not follow the standard, now you have to think about the transformation that takes you from your world to the physics world. Can get annoying when you want to fix a bug. (b) When you want to share code with someone, or bring someone over to help with development, they have to get used to your convention. (c) When using standard 3D models, you always have to have a transformation above them to prevent them from looking sideways.
Now to add to question 2, it is sooo useful to think of X, Y, and Z as not just three letters, but as right, up and backwards. Every character in the scene has a local coordinate system attached to them, and in their local coordinate frames X is always right, Y is up and Z is backwards. Once you have this, now you can make sense of vectors that you print out, or write your algorithms in a way that makes sense. Let's say you have two characters A and B, and you want to do something if one of them is facing the other. You can simply find B's location in A's coordinate frame (Ta^-1 * p_b), look at the vector you get and see if Z(backwards) is negative and X(right) and Y(up) are small, because that vector tells you how much backwards, right and up B is with respect to A.