See issue504

Confirmation: it works nicely on the two machines I tested. As discussed off the
tracker, for the future I would suggest *only* having library environment
variables that are specific to a certain bit width. I think this would eliminate
a potential source of errors and would simplify the documentation because there
would be fewer different variations to document. But this is a lower priority
item for me than this issue here was.

Merged and updated the documentation.

Patch looks good to me. Feel free to merge, I'll try to test it as soon as
possible. Once it is merged, can you also update the LP build instructions page?

I created a pull request on Bitbucket:
https://bitbucket.org/FlorianPommerening/downward-issues/pull-requests/25

The change duplicates some code from FindOSI.cmake but CMake find-scripts should be 
self-contained, so I didn't want to factor this out.

My 32 bit and 64 bit version of CPLEX are installed in the same directory, but it 
looks like its working correctly from some debug output I introduced for testing 
locally.

We introduced separate variables for different build environments for COIN/OSI but 
not for CPLEX in issue504. Now, we also want to add them for CPLEX.