Move tensor encoding/decoding into the Index class

[kylebgorman]

This is a notice of my plans to move the encoding methods (which take strings and make tensors) and decoding methods (which convert tensors back into strings) into the Index class itself. Currently this ability lives inside of the DataSet class and I really don't know why it belongs there. This should substantially reduce weird coupling, like for instance what happens here with all those chained method calls.

As part of this, the Index class will have two class methods, one which reads the index from a .pkl file, and one which constructs the index from a list of file handles. This latter should give me a place to compute max tensor sizes, as needed for #50.

I also think the index will then become a better object to unit test.