Single element assemble + override default assemble operation

[ahojukka5]

So, if we introduce "local buffer", containing workspace/memory for assembling one element without any additional memory allocations, we could go back to assemble one element at a time.

It would be cool if we can make dispatching work so that we can override or extend assembling of a single element if we have some special case we need to take care of. Would it work if we have something like:

function assemble_stiffness_matrix!(problem, assembly, other_stuff, Val{:E1})
    # my special assembly procedure
end

where :E1 is the name of the element.

We also should construct everything such that we can assemble mass matrix, stiffness matrix, force vector and so on separately or all together.