Create a new pass to convert the Pi groups computed by dimensional matrix analysis into a canonical form

[phillipstanleymarbell]

Create a new pass to convert the Pi groups computed by dimensional matrix analysis into a canonical form and to weed out duplicate kernels:

• Use the lexicographic ordering (or some hash) of the dimension names to implement a canonical "polarity" (e.g., the dimension with the lowest lexicographic index goes into numerator, hence its row in the kernel should always be positive, so always multiply the kernel by the row polarity of the lexicographically smallest dimension.
• Once in this canonical form, remove Pi groups (not individual kernels columns, but entire kernel groups) that duplicate existing ones after canonicalization.

This pass would ideally be run before the pass being implemented by #358 and #359 (and ideally also run before #354). It however need not be implemented immediately since its effect is not needed for correctness of #358 or #354.

This would likely be best done using Eigen, so I can see its implementation residing in newton-eigenLibraryInterface.cpp, e.g., after newtonEigenLibraryInterfaceGetPiGroups(), but being called from a new irPass file (e.g., newton-irPass-dimensionalMatrixPiGroupCanonicalization.c), in much the same way newton-irPass-dimensionalMatrixPiGroups.c is just a wrapper to call newtonEigenLibraryInterfaceGetPiGroups. This will allow us to also cleanly enable it or not using flags.

[Siegfriedchao]

This issue has been further enhanced/modified, see #383 and #384. This pass should now be able to sort the labels of parameters lexicographically and return a reordered null space. The weed out has now been moved to #384.

[Siegfriedchao]

Typo in commit 9071a56: should be "moving on to #383 and #384"

[phillipstanleymarbell]

This is merged. Moving to verify.