DenseMatrix functionality improvements

[harakas]

After a slight brush with DenseMatrix, here's some stuff I found not working that well:

• DenseMatrix is missing __sub__, currently can do A+B, but not A-B.
• DenseMatrix does not support left side multiple with a Basic, that is Symbol('x')*A fails. Add __rmul__ or modify Basic?
• Adding and multiplying of DenseMatrix objects of incompatible size should result in Python errors/exceptions, not crashes.
• Indexing of a single column or row matrix should work with a single index, like A[1].
• Add vector specific methods like dot, norm, cross (I suppose this should be done on the symengine/C++ side actually first).

[harakas]

• It would also be nice to have sympy style initialisation working, like A = DenseMatrix([[1, 2, 3], [4, 5, 6]]).
• This should also allow one to create single column vectors as v = DenseMatrix([1, 2, 3]).
• Consider abandoning the entire C++ style initialization.
• Currently one can do A+5, which results in 5 being added to every cell. I think this is something that should not be allowed. There is add_scalar already, no need to muddle the algebra in this way.

[certik]

@harakas thanks a lot for your feedback. We should definitely improve the Python interface.