Why is the Manhattan distance (or block distance) appropriate when I have a discrete data set and the Euclidean distance is appropriate when I have continuous numerical variables?
I don't think the appropriateness depends on the discrete / continuous nature of the data. (In my opinion the Manhattan metric makes sense when the move directions are limited to North-South and East-West (or similar axes).) In fact when there are more move directions allowed we get different metrics, all way to the limit when we get unrestricted move directions, and Euclidean distance.