Chunk - Optimise Flatten implementation.

The current flatten performs one Chunk.Queue insertion
per source chunk, which means it creates a new
"Chunk.Queue" object and one "Queue" object for each one.
Instead, we can avoid those objects and use a Queue- builder.

core/shared/src/main/scala/fs2/Chunk.scala

@@ -1007,7 +1007,8 @@ object Chunk
     *
     * This is similar to a queue of individual elements but chunk structure is maintained.
     */
-  final class Queue[+O] private (val chunks: SQueue[Chunk[O]], val size: Int) extends Chunk[O] {
+  final class Queue[+O] private[Chunk] (val chunks: SQueue[Chunk[O]], val size: Int)
+      extends Chunk[O] {
 
     private[this] lazy val accumulatedLengths: (Array[Int], Array[Chunk[O]]) = {
       val lens = new Array[Int](chunks.size)
@@ -1221,9 +1222,15 @@ object Chunk
         if (ffa.isEmpty) Chunk.empty
         else if (ffa.size == 1) ffa(0) // short-circuit and simply return the first chunk
         else {
-          var acc = Chunk.Queue.empty[A]
-          ffa.foreach(x => acc = acc :+ x)
-          acc
+          val sqb = SQueue.newBuilder[Chunk[A]]
+          var totalSize = 0
+          ffa.foreach { (ch: Chunk[A]) =>
+            if (!ch.isEmpty) {
+              sqb += ch
+              totalSize += ch.size
+            }
+          }
+          new Chunk.Queue[A](sqb.result(), totalSize)
         }
 
       override def traverse_[F[_], A, B](