Update the Julia interface of ColPack

[amontoison]

• Recompiled version of ColPack without OpenMP. It seems to fix the non-determinism.
• Add more options for order and method in the C++ code like "NAIVE_STAR" or "DYNAMIC_LARGEST_FIRST".
• Reimplement the C++ class Timer such that it works and that we have highly accurate timers directly from ColPack for the ordering and coloring.
• Add a folder gen to automatically regenerate the Julia wrappers in libcolpack.jl from the header cinterface.h if the C interface of ColPack is updated.
• Provide three new routines to get the ordering from ColPack as well as timers for coloring and ordering.

Comments:

• Even in the serial version of ColPack, we still have some minor bugs. The ordering LARGEST_FIRST returns invalid ordering that is not a permutation is some rare cases. The method TRIANGULAR fails with segmentation fault.
• The interface for CSC matrices that I implemented in ColPack seems to provide some errors sometimes, I have an idea of what is the issue but I don't think that it is relevant to spend too much time on it. We can do the conversion CSC -> CSR -> ADOLC format in Julia.
  The ADOLC reader was not modified by my commits.
  The CSR should work but it was buggy before and I fixed it, so for the current paper it is more relevant to use as much as possible things not added by me in the C++ code (ADOLC format / MTX files).

[gdalle]

Looks good to me, I presume the timer tests are deactivated because timers are not initialized?
Can we add some tests for the new ordering retrieval, then this should be good to go?

[gdalle]

Once this is merged we could do a new release of ColPack.jl? If I understand correctly, we now go through the ADOL-C format systematically, which might fix #21?

[amontoison]

The format ADOL-C has been used for a long time in ColPack, and it's quite safe to use.
The other formats, CSR and CSC, are less tested, and I found a few bugs.
Even though I fixed them in my fork, we should play it safe.
If issue #21 is fixed, it's because ColPack was recompiled without OpenMP; it's not related to the graph's input format.

[gdalle]

At the beginning of #21 we had also found a case where the binary was deterministic but the coloring through CSR/CSC was random, that's what I was referring to here