Clean up hard-coded loop bounds

[dmcdougall]

Opening a ticket for dicussion/todo item.

The character of bugs such as the one fixed by #1459 can easily be avoided by not duplicating hard-coded constants. If QUDA is willing to chew a C++17 requirement, I think one can use std::size to retrieve the size of the input array arguments, store it in a variable, and re-use the variable.

[maddyscientist]

We already have a C++17 requirement, so that's not a problem. Absolutely we should be removing hard-coded constants, though in general I think these quantities should be derived from parameters, rather than queried from size(). E.g., in #1459, the quantity 9 comes from nColor * nColor where nColor = 3.

Anyway, leaving this issue open, and we can close this when we eventually get around to this....