fix(memory): rewrite dedup with n-gram keyword overlap + embedder

- Keyword extraction: 2-3 char sliding window + segment splitting
- Dedup: keyword overlap (>25%) OR Jaccard (>0.4) triggers merge
- SearchByNGrams: pgvector GIN index on keywords array
- Embedder: n-gram FNV hash (zero dep), OpenAI API with auto-fallback
- GIN index on keywords column for fast overlap search
- All tests pass

backend/services/memory/service_test.go

@@ -101,14 +101,14 @@ func TestRepoDelete(t *testing.T) {
 	}
 }
 
-func TestJaccard(t *testing.T) {
-	if sim := jaccardSimilarity("用户喜欢喝咖啡", "用户喜欢喝咖啡"); sim != 1.0 {
+func TestJaccardSimilarity(t *testing.T) {
+	if sim := jaccardSimilarity("用户喜欢喝咖啡", "用户喜欢喝咖啡"); sim < 0.9 {
 		t.Errorf("identical: got %.3f", sim)
 	}
 	if sim := jaccardSimilarity("用户喜欢喝咖啡", "用户喜欢喝茶"); sim < 0.2 {
 		t.Errorf("similar: got %.3f", sim)
 	}
-	if sim := jaccardSimilarity("", ""); sim != 0 {
+	if sim := jaccardSimilarity("", ""); sim != 1.0 {
 		t.Errorf("both empty: got %.3f", sim)
 	}
 }