Improved a slight inefficiency when reallocating

[9x14S]

The inefficiency lies at the repeated calling of realloc() to make space for each character read just for the buffer to be minimized at the end.

[dmalan]

This is deliberate, actually. We used to allocate more bytes at a time but decided to simplify the code, particularly since the human is the bottleneck with get_string, not the underlying system calls.

As written, too, looks like this PR risks overflowing capacity, fyi!