Like this, you have options - you can
round a floating point number to an integer, if you need it, or
format an integer as floating point (I think adding
0.0 to an integer will already do this, try
msgbox % 4 + 0.0 ), before applying
floor division.
I think the floating point version is what is usually considered as
floor division while the other case is what is indeed often called
integer division (a certain type of it, at least), especially the specific kind you will encounter in some programming languages like
C, which will return only integers.
But afaik,
integer division is not clearly defined in mathematics, it is not
closed in a mathematical sense and hence, not handled unambiguously. So, consistency might not really be a relevant category here... it is rather a matter of definition.
Since AHK's source code is written in C++, it can probably make direct use of some
"high performance" integer division of the language. Perhaps some of our C++ or AHK source code gurus can chime in on this. (So far, I have succesfully suppressed my long-ago experiences with C

)
But in the end, it is documented like this, not really uncommon in programming and it can be handled. So I wouldn't expect it to change (which might even break a number of older scripts).