Trait for 1d to nd index conversion

Start with the generic shuffle example and inspect the Array, Array1, and Dimension indexing, flattening, and reshaping APIs mentioned in the issue. A complete contribution would establish whether a single usize can be converted safely for arbitrary dimensions and document or test the resulting behavior.

Given an array of arbitrary dimension, is it possible to index into it using a single usize?

fn shuffle1<E>(values: &mut Array1<E>) {
    let mut i = values.len();
    while i >= 2 {
        // invariant: elements with index >= i have been locked in place.
        i -= 1;
        // lock element i in place.
        values.swap(i, thread_rng().gen_range(0, i + 1));
    }
}

fn shuffle<E, D: Dimension>(values: &mut Array<E, D>) {
    let mut i = values.len();
    while i >= 2 {
        i -= 1;
        values.swap(i, thread_rng().gen_range(0, i + 1));
        // This won't work as a single usize is not necessarily a valid index,
       // the usize must be converted to a index of dimension D
    }
}

One possible solution is probably to flatten the array, then shuffle it, then reshape it back, although I do think a method for indexing into the array with a single integer could be useful in general.