Use estimate of row count when growing arrays

[davidanthoff]

Right now we just grow them by a fixed factor. Instead, we could compute the average row size in bytes, plus standard deviation or something, and then grow based on the remaining bytes in the file.