…templates.

Preliminary fix for #142

Instead of calculating degeneracies first, find reaction templates after
generating reactions, then combine degenerate reactions only if reaction
templates match. These isomorphic reactions remain separate
(unlike in RMG-Java) in case someday we want to calculate the separate
transition states.  The reactions are added to the core and printed
to chemkin as duplicates.

Issues still remaining:
- How to identify the correct reverse reaction when having multiple TS's.
 My solution is to match the correct reaction pairs- but this only works in
the case of a bimolecular A+B=C+D reaction.  If the reaction is unimolecular,
the reaction pairs will be the same and the correct path cannot be distinguished.
- If kinetics are identical, they should be combined through degeneracy, in this
commit they remain as duplicates.

If forward reaction has the same template, then do not include it; if not, then
we have found an additional transition state and the reaction should be included.
Do not include any reverse reactions if the family has its own reverse otherwise
we are including kinetics for more than one direction.

…This is faster.

…to figure out

which reverse reaction (if there are multiple transition states) matches the forward reaction.

This is the best I can do for now.  Maybe if there is a way to look at the mappings
of the labeled atoms this can be done better.  Already we are accounting for multiple TS's
in the forward direction, so that is already better than before.