dependencies on outlier detection library

[feamster]

The package depends greatly on some lesser-used (and less useful) outlier detection functionality, which uses the pyod library. There are two possible paths forward here:

• eliminate the dependency entirely if most of the outlier detection functionality is in sklearn
• split this library into two, where the outlier detection functionality is a separate (sub?)-module of netml itself

@jesteria happy to hear your thoughts on this. Based on what we decide, happy to work on implementing it.