Switch to 64-bit indices

[mlubin]

Closes #41

Tests are passing with Soplex and SCIP. I didn't actually try an instance that exceeds 2 billion variables/constraints.

TODOs:

• [ ] The interface with HiGHS should be tested. It's probably broken.
• [x] Is the HitCount typedef in Sparsify.hpp safe as an int16?

Before merging, I recommend:

• Checking the changes very carefully
• Running benchmarks (are there standard ones?) to check if there's a performance loss in common case where 32-bit integers would be sufficient

Minor:

• The code calling LUSOL might not need to make an extra copy of the matrix anymore.

[lgottwald]

The hit count is safe as rows are not tried when they exceed the types length

[mlubin]

The hit count is safe as rows are not tried when they exceed the types length

Thanks for clarifying. I will test HiGHS next, although given the holidays, it may be a week or so before I get to it.

[mlubin]

I'm not able to test HiGHS because it appears that PaPILO isn't compatible with HiGHS master:

PaPILO/src/papilo/interfaces/HighsInterface.hpp:84:22: error: use of undeclared identifier 'OBJSENSE_MINIMIZE'
      model.sense_ = OBJSENSE_MINIMIZE;

papilo/interfaces/HighsInterface.hpp:89:13: error: no member named 'nnz_' in 'HighsLp'
      model.nnz_ = consMatrix.getNnz();

Nevertheless I think the PR is ready to be reviewed.

[lgottwald]

I already looked at the changes and it looks ok. The HiGHS interface needs updating independently of this anyways as some things on the HiGHS side changed since I wrote the interface. Currently performance tests of this change are being conducted and once they show that this does not hurt performance significantly we can merge this.