Can add vector to 1xn matrix

You can in Julia, can't in MATLAB.

[StefanKarpinski]

Is this a bug or what?

[JeffBezanson]

Just a difference from matlab to consider.

[ViralBShah]

This whole equating of vectors to a 1xn or nx1 matrix is going to create more such issues. Do we have a view on this? I am inclined to keep it simple (which may make matlab users unhappy) for the first release.

[JeffBezanson]

We can fix this and #152 in one pass by adding size checks.

[StefanKarpinski]

We still need a policy that will be

a) not annoying
b) not introduce confusing edge cases like the ones that Matlab is full of

Between these two, I'm starting to reconsider the option of filling in an arbitrary number of trailing dimensions of size 1. That would automatically make vectors and column matrices the same thing, and 1x1 matrices the same thing as scalars. It really, really messes with the type system though.

Maybe less drastic than that, we can just automatically fill in trailing ones for parallel ops like +. In that case, vectors and column matrices could just be added, but not vectors and row matrices. Which seems acceptable.

I also think that it makes sense to allow mxn ./ mx1 and replicate the 1 dimension of the second matrix. This makes it much easier to take row-normalized and column-normalized matrices without unnecessarily making huge, repetitive matrices of normalization factors.

[ViralBShah]

For now, commit 7a8960f fixes this bug.