Use N or N-1 instead of empirical kmax in degree histogram

Tweaks `DegreeDivergence` to use N (if there are self-loops) or N-1 (if there
are not) instead of the largest observed degree when constructing the degree
histogram. This works because a `Counter` behaves like a `defaultdict(int)`, so
the list comprehension

```
hist = np.array([counter[v] for v in range(max_deg)])
```

pads out the zeros automatically. It should rarely affect results, but has the
advantages of (i) being slightly more accurate for the pedants among us and (ii)
being faster for some reason. And while it doesn't affect the JSD, if we ever do
implement #174, maybe it ends up mattering somewhere else.

netrd/distance/degree_divergence.py

@@ -48,7 +48,10 @@ def degree_vector_histogram(graph):
 
             """
             vec = np.array(list(dict(graph.degree()).values()))
-            max_deg = max(vec)
+            if next(graph.selfloop_edges(), False):
+                max_deg = len(graph)
+            else:
+                max_deg = len(graph) - 1
             counter = Counter(vec)
             hist = np.array([counter[v] for v in range(max_deg)])
             return vec, hist