Smart chunking about pop

[tchardin]

@gallexis yeah we use that lib for detecting the type when posting to the HTTP server so that should work fine.

[gallexis]

I'm considering using this lib : https://github.com/gabriel-vasile/mimetype to detect mimetype using the first 512 bytes of a file.

Using the lib should give us more accuracy at the price of adding some slight delay when importing a file (need to be benchmarked).
We could also use both approch, where we first try based on extension name, if none is given, we use the library.

[gallexis]

@tchardin chunk.NewBuzhash(r io.Reader) only takes a reader, I can't set its chunk size to 16kb, is it ok like that?
https://github.com/ipfs/go-ipfs-chunker/blob/master/buzhash.go#L24