Support non-binary word vectors in compute-accuracy

[Witiko]

This PR makes adds the binary and text format identifiers to the header of word vectors to tell the difference between binary and non-binary word vector formats. The compute-accuracy script has been extended to recognize both binary and non-binary word vectors. Binary word vectors without the format identifier are also recognized to remain backwards-compatible.

The PR has been successfully tested as follows:

• Three word vector files were produced using -binary 0 on HEAD, -binary 1 on HEAD, and -binary 1 on HEAD^.
• All three word vector files were evaluated using compute-accuracy to see if the standard deviation of the total accuracy was below 1%.

[Witiko]

I am sorry, I did not realize the vector file format was a de-facto standard. I suppose a better course of action might be to add a -binary argument to compute-accuracy.c instead.