* the following strategies:
 *
 *  1. If the current call already has a connection that can satisfy the request it is used. Using
 *     the same connection for an initial exchange and its follow-ups may improve locality.
 *
 *  2. If there is a connection in the pool that can satisfy the request it is used. Note that it is
 *     possible for shared exchanges to make requests to different host names! See

 * </ul>
 *
 * <p>Advantages compared to {@link com.google.common.collect.ImmutableList ImmutableList}{@code
 * <Long>}:
 *
 * <ul>
 *   <li>Improved memory compactness and locality.
 *   <li>Can be queried without allocating garbage.
 *   <li>Access to {@code LongStream} features (like {@link LongStream#sum}) using {@code stream()}
 *       instead of the awkward {@code stream().mapToLong(v -> v)}.

 * </ul>
 *
 * <p>Advantages compared to {@link com.google.common.collect.ImmutableList ImmutableList}{@code
 * <Double>}:
 *
 * <ul>
 *   <li>Improved memory compactness and locality.
 *   <li>Can be queried without allocating garbage.
 *   <li>Access to {@code DoubleStream} features (like {@link DoubleStream#sum}) using {@code
 *       stream()} instead of the awkward {@code stream().mapToDouble(v -> v)}.

 * </ul>
 *
 * <p>Advantages compared to {@link com.google.common.collect.ImmutableList ImmutableList}{@code
 * <Integer>}:
 *
 * <ul>
 *   <li>Improved memory compactness and locality.
 *   <li>Can be queried without allocating garbage.
 *   <li>Access to {@code IntStream} features (like {@link IntStream#sum}) using {@code stream()}
 *       instead of the awkward {@code stream().mapToInt(v -> v)}.
 * </ul>

    // The Dolphin algorithm is attractive because it does the fewest array reads and writes: each
    // array slot is read and written exactly once. However, it can have very poor memory locality:
    // benchmarking shows it can take 7 times longer than the other two in some cases. The other two
    // do n swaps, minus a delta (0 or 2 for Reversal, gcd(d, n) for Successive), so that's about