1. Then how do games using directinput tell the difference between qwerty and azerty?
Usually by going through extra steps to trigger the translation. But they only tend to do this when the user is typing text, or when they want to display something recognisable to represent a key. Other times the translation would get in the way - if the default movement keys are WASD on a US keyboard, it doesn't care that a FR keyboard has different legends.
2. I don't see how this is entirely relevant anyway because irregardless of the cause there still remains the fact that these things happen and conflicts are run into if a user attempts to use things on the numpad block. The thing with the enter key is the first and only time I have ever seen it not respond as expected. The original point was that an arrowkeyless (and home clusterless) keyboard would absolutely 100% positively have to send out the keycodes that the discreet keys use instead of the potentially troublesome ones that they currently do. Delving into the hows and whyfores of what is causing the issue of those scancodes to not work is academic.
Sorry to be blunt, but this is just FUD. I know from experience that these scancodes work perfectly well in practice, and the reality is that it is no more a problem than using a tenkeyless (less, probably, but there's not much point in trying to quantify that, it depends so much on what programs someone uses). I know this because I've been using them exclusively for years.
On Windows, anyway. Other OS, not so sure. I already said a modern take on the 84-key AT layout should be able to send the discrete key's codes. Heck, it could have as many modes as it wanted for the numpad!
I think if I were designing such a keyboard, it would have two keys instead of the double width insert, and three modes: standard numpad using standard codes, and two using the discrete keys' codes - one laid out the same as the numpad, and one laid out like the discrete keys.
But that doesn't detract from the fact that using an old 84-key AT keyboard on Windows today is perfectly viable as far the numpad cursors etc are concerned. (The lack of F11 and F12 is the real problem).