A 'balanced' version of sequencing, which realistically avoids stack overflow errors for any finite Applicative. The Applicative laws (specifically associativity and right/left identity) ensure this yields the same result as the usual right-fold based version.

BSequence.scala

trait Applicative[F[_]] {
  
  /** 'Balanced' sequencing, to avoid SOE. */
  def bsequence[A](ms: Seq[F[A]]): F[IndexedSeq[A]] = {
    if (ms.isEmpty) point(Vector())
    else if (ms.size == 1) ms.head.map(Vector(_))
    else {
      val (l,r) = ms.toIndexedSeq.splitAt(ms.length / 2)
      apply2(bsequence(l), bsequence(r))(_ ++ _)
    }
  }
  
  /** Also sometimes called `map2` or `lift2`. */
  def apply2[A,B,C](fa: F[A], fb: F[B])(f: (A,B) => C): F[C]
}