Fix wordnet's all_synsets() function

[ekaf]

Fix #3077, and a few related problems in wordnet.py.

1. With the English Wordnet, properly distinguish satellites from adjectives:

from nltk.corpus import wordnet as wn
print(wn.get_version())

3.0

print(len(list(wn.all_synsets(pos="a"))))
18156

print(len(list(wn.all_synsets(pos="s"))))
10693

2. With other languages, a part of the problem is due to the synset_from_pos_and_offset() function, which returns an empty synset object when an offset is incorrect. This leads to inflated synset counts with OMW for two reasons: OMW does include some incorrect offsets, and with later wordnets some synsets that are lost in mapping should not be counted. So, when a synset does not exist, this PR returns None instead of an empty synset object:

ss = wn.synset_from_pos_and_offset('n',1234)
print(ss)

None

wn.add_omw()
print(len(list(wn.all_synsets(lang="hrv"))))

23115
print(len(list(wn.all_synsets(pos="n", lang="hrv"))))
16177
print(len(list(wn.all_synsets(pos="v", lang="hrv"))))
4736

Another problem is that the handling of satellites in the mappings was inconsistent. The mapping procedure has now been rewritten in a style that makes it easier to follow the underlying algorithm.

from nltk.corpus import wordnet31 as wn31
print(wn31.get_version())

3.1

wn31.add_omw()
print(len(list(wn31.all_synsets(lang="hrv"))))

23098
print(len(list(wn31.all_synsets(pos="a", lang="hrv"))))
1363
print(len(list(wn31.all_synsets(pos="s", lang="hrv"))))
444

[tomaarsen]

I've added some simple tests. I also accidentally brought in 2f34b52 into this PR, so I rebased and removed some of the unnecessary commits. Apologies for this. Make sure to update locally to avoid issues if you make changes!

I had some comments and requests for changes. I hope you'll be able to have a look at them.

[tomaarsen]

This is looking good! Thanks for always helping maintain Wordnet in NLTK!