Remove (or fix memory usage of) strassen multiplication

[OlivierSohn]

While working on #56, I saw that for every testable matrix size, multStrassen is /much/ slower than any other multiplication method. Also, for sizes > 256, the benchmark cannot finish because the memory used is too much (I saw the binary hang and use > 4GB).

Maybe there is a space leak somewhere that should be fixed, or if the algorithm is inherently slow, it could be removed from the API to not confuse the users.

[OlivierSohn]

Maybe a way to fix memory usage could be to implement this