Tpetra: Split makeOptimizedColMapAndImport test into <= 4 vs. > 4 MPI procs cases

[mhoemmen]

Tpetra's current test for makeOptimizedColMapAndImport requires 5 MPI processes. Default value for MPI_EXEC_MAX_NUMPROCS is 4, so the test does not run by default. I would still like to keep the tests that require more than 4 processes, however. Thus, we should add more tests that require <= 4 MPI processes, and split the test into the tests that need <= 4 MPI processes, and those that need > 4 MPI processes.

See discussion starting here: #2198 (comment)

@trilinos/tpetra

[mhoemmen]

I've been working on this, and found that makeOptimizedColMapAndImport (not the new Import constructor) might be broken. It reorders all the output (optimized) column Map entries, not just the remote ones. I'm debugging now.

[mhoemmen]

Thanks @william76 ! :D