>>This has less (or no) collisions http://msdn.microsoft.com/en-us/library/system.security.cryptography.sha256(v=vs.110).aspx
>
>Do you know why this one is a better approach to have a better success at detecting duplicates?

(1) Because the hash is larger in size
(2) Because here http://en.wikipedia.org/wiki/Cryptographic_hash_function look at the column collisions
I suspect they are SHA-2

If you want SHA-3 - looks like it's in here - http://hashlib.codeplex.com/


>
>Also, there is a line which puts the pointer at the beginning of the file. Isn't by default an initialization of a stream to have its pointer at the beginning of the file?

Yes