Python algorithms and related VectorFst methods have inconsistent return values (self or None)

[dhdaines]

Some of them modify the FST in-place, others don't and return a new FST. This is not at all reflected in their return values - some in-place operations return self while others return None.

If some in-place operations return self, then they should all return self so that they can be chained, particularly since some of them which currently return None are non-destructive. These are the ones that should return something and currently don't:

• tr_sort (really important, so you can do fst1.tr_sort(False).compose(fst2.tr_sort(True)) for instance)
• rm_epsilon (not an in-place operation either! the version in algorithms does return a new FST) SEE BELOW
• tr_unique

Also the in-place nature isn't documented, but that is fixed by #278

[dhdaines]

Note that #278 would have to be updated with these changes ... I'll wait for a review, if you approve both of these then I can merge them into a new PR.

[dhdaines]

Worse yet rm_epsilon actually is an in-place operation, but the version in rustfst.algorithms.rm_epsilon returns ... something else?