Rethink triangular matrix handling

[Ulfgard]

the only thing distinguishing those two is how they are stored in the backend, i.e. this is a pure implementation detail which should not have an effect on the default implementation - they should be exactly the same. also there should not be any logic in the expression classes regarding this distinction

[Ulfgard]

i just realized that this is not the case - at least for the block implementation which can make use of the dense storage. Still, it feels wrong to have a distinction on the level of expressions.

[Ulfgard]

one example of where this fails is when we have C=A % T where T is triangular. The kernels are perfectly capable of doing this via trans(C) = trans(T) % trans(A), however, on expression level this logic is not implemented, so this will fail.