This problem requires two parts: a way to turn an arbitrary number into a
numeric string, and locating the indices of a specific letter in a char buffer
that dynamically grows. It appears the second part depends on the first part for
this problem.

The first part is made more difficult by the fact that it must be ordinal (use
"first" instead of "one", and so on). Perhaps this could be made easier if the
number was converted into a list of tokens that can each be directly converted
to their string equivalent. This method makes it easy to make it ordinal since
it can be simply done by making the last numeric token ordinal.

In English, it may be easier to convert the number into each "thousand". For
example, "123456" could be split into "123 thousand", and "456". They can then
be subdivided into tokens that can represent a single numeric word.

"123 thousand" -> "100", "20", "3", "thousand"
    -> onehundred twenty three thousand
"456" -> "400", "50", "6"
    -> fourhundred fifty sixth