Faster LinkedHashSet head()

While writing up vavr-io#2727 I noticed that `LinkedHashSet.head()` is implemented
`iterator().head()`. This is inefficient because `queue.iterator().next()`
makes a call to `.head()`, but also to `.tail()` so as to prepare for the
next `head()` call. Given the structure of the underlying `Queue`, the
`head()` call is worst case `O(1)` but the tail call is worst case
`O(n)`. The present worst case will be achieved if there have never been
overwrites or removals from the set, which is probably a fairly common
case.

src/main/java/io/vavr/collection/LinkedHashSet.java

@@ -631,12 +631,12 @@ public T head() {
         if (map.isEmpty()) {
             throw new NoSuchElementException("head of empty set");
         }
-        return iterator().next();
+        return map.head()._1();
     }
 
     @Override
     public Option<T> headOption() {
-        return iterator().headOption();
+        return map.headOption().map(Tuple2::_1);
     }
 
     @Override