Add generic vector/matrix operations

[vbaconnet]

Overloading some operators for the vector_t and matrix_t classes, and add some additional math routines like cadd2, device_cadd2 (a(i) = b(i) + c) and device_add3 (a(i) = b(i) + c(i)).

This is to facilitate the usage of vector_t and matrix_t especially with GPUs. This should not break anything since it's just adding to what we have, only thing I changed is the intent on some arguments in sub3 and add3, @njansson will this be a problem?

Tested on cpus, nvidia, and amd gpus. Not tested with OpenCL.

Contrary to the assignment operator where if we do v = w and v%x is already allocated, we free v and re-initialize it to have the same size as w, all the other operations assume that if one does v = a + b and v is already allocated then v should have the same size as b and a. So there is no implicit reallocation except for the assignment operator.

[vbaconnet]

Not sure I understand what is wrong with the checks, if someone could take a look that would be much appreciated :)

[njansson]

I would say we should merge this, and I can take upon me to fix the generic interface (+ ensure correct inlining in key kernels)