Implement missing OpenMP kernels

[tcojean]

The purpose of this issue is to have a full OpenMP implementation of the kernels before the Ginkgo release 1.0.0.

Here is @hartwiganzt's original comment on this issue:

On the OMP issue: there are in total about 25 NOT_IMPLEMENTED in the omp branch.
I would say: let's copy complete those - in the worst case with the reference
implementation and some note saying that this still needs to be parallelized.

One way to do this is to copy the reference kernels and add some simple parallelization when possible. When doing so, it is very important to add a note in the README that the OpenMP implementation is present but needs major improvement, or is not fully parallelized (depending on the case).

On a related issue, two OpenMP tests seem particularly long. The COO test is definitely very long (in valgrind it reaches the timeout of 1500 seconds, whereas the others take around 200 seconds!). The CGS solver test also seem to possess some issues, but not to the severity of the COO one.

[tcojean]

Closing this as this is now either done, or tracked by #286