Peter M. Groen / openbr

Rscript will actually crash partway through writing the output pdf, presumably
because the recognition curves are degeenrate in this case.

Address #133

By default, when Transform::fromAlgorithm is used, attach a stream transform
as the root of whatever transform is created (can be disabled with a flag).

Actually implement stream project(template), give a warning if only returning
one of N templates.

If we get a 1 column matrix in eval, check if the query and target sets are
of matching size, and construct a pairwise mask for that matrix.

Add support for doing eval in terms of a series of pairwise comparisons

We do need to do a transpose, but the inital matrix should have columns and
rows reversed, so that the matrix side remains correct.

Similar to the read methods, cv::mat contains type information, so there is no
particular need for separate methods for masks and similarity matrices.

Also, fix an apparent bug where .mat files read by openbr were transposed
(compared to their dimensions when writing the files).

Since .mtx files include a matrix type in their header, we don't really need
separate functions to read similarity matrices and masks.

This drops special case support for creating identity matrices via readSimmat
and readMask (since in that case, there is no actual header to read to get a
matrix type).

Fix compile under visual studio