Scikit-fingerprints

Breaking changes

- changed `ConformerGenerator` error handling argument; the default behavior should stay the same
- fixed PubChem SMARTS patterns

New features and improvements

- new fingerprints: MACCS count, MQNs
- better looking docs
- docs for fingerprints: autocorrelation, MACCS, PubChem, Topological Torsion
- some fingerprints were missing the `batch_size` argument, fixed

New features and improvements

- new fingerprints: functional groups
- docs: do not display the whole path to classes
- fix imports for older versions of scikit-learn

Breaking changes

- reduced default number of `max_attempts` in `ConformerGenerator` from 10000 to 1000, to greatly increase the speed for most use cases

New features and improvements

- new fingerprints: USR, USRCAT

Breaking changes

- add parameter constraints in scikit-learn style for parameter validation
- heavily refactored `ConformerGenerator` to allow generation of multiple conformers
- MAP fingerprint now requires generated conformers, instead of generating them inside

New features and improvements

- new fingerprints: Ghose-Crippen, Klekota-Roth, Laggner, Layered, Lingo, Mordred
- add heavy atom pair scaling to Atom Pair fingerprint
- docs for fingerprints: Atom Pair, ERG
- fix MAP fingerprint bugs, improve stability and speed
- reduce E3FP logs
- add Makefile
- refactor time benchmark