Add possibility to detect encoding from input streams

Right now it seems the detector is only able to work with input arrays, it would be very useful to allow input streams as well. This makes sense when the input files are very large and it's not memory efficient to load them entirely into memory.

[JinsukKim]

Thanks for the input and sorry for belated response. Would appreciate your contribution to make it better. The way the detector works however doesn't require too much memory. It was mainly designed with web documents in mind, and first few KB will suffice to make a probabilistic decision. It also returns the # of bytes it consumed to do that. It rarely goes over 3~4 KB. The rest of the input is ignored.