[feature] add multithreading support

Almost all dilate/erode ( with common SE:diamond,box,...) algorithms could be parallelized 
By spltting the image into subsets,mapping each subset  to an appropriate kernel through a thread pool


