Fix zero-length norm bug with RM3 (#376)

Avoids zero-length feedback documents, which causes division by zero when computing term
weights. Zero-length feedback documents occur (e.g., with CAR17) when a document has only 
terms that accents (which are indexed, but not selected for feedback).

src/main/java/io/anserini/rerank/lib/Rm3Reranker.java

@@ -165,7 +165,12 @@ private FeatureVector estimateRelevanceModel(ScoredDocuments docs, IndexReader r
     for (String term : vocab) {
       float fbWeight = 0.0f;
       for (int i = 0; i < docvectors.length; i++) {
-        fbWeight += (docvectors[i].getFeatureWeight(term) / norms[i]) * docs.scores[i];
+        // Avoids zero-length feedback documents, which causes division by zero when computing term weights.
+        // Zero-length feedback documents occur (e.g., with CAR17) when a document has only terms 
+        // that accents (which are indexed, but not selected for feedback).
+        if (norms[i] > 0.001f) {
+          fbWeight += (docvectors[i].getFeatureWeight(term) / norms[i]) * docs.scores[i];
+        }
       }
       f.addFeatureWeight(term, fbWeight);
     }