minimal edits to get faster fragmenting

[pstjohn]

@IchiruTake, do these edits come close to what you've implemented without completely changing the fragment API? My timing comparisons have them as fairly close

[IchiruTake]

@pstjohn Yes, if you said by algorithm it is matched but by your algorithm, we are lacking control of the molecule. The class molecule make me nervous of multiple states across. If I add both smiles and mol that are not the same, they are still valid either.

Moreover, I have extensively improved the documentation to make sure for further customization. Take the 5 status of the bond I opened in the class MolProcessor or the updated documentation of BDFE in predict.py

It would be great to have stricter control and a systematic approach for what user should do rathen pushing every possibilities.

Overall, this PR has done it great tuning job, but lack of further feature and variable control for beeter memory management.

[IchiruTake]

I have reviewed your fragment.py which allowed further molecule control