Allow for adding the entire Batch as one tensor

[wvabrinskas]

right now we iterate over all instances of the batches and perform a forward operation all at once. If we add a 4th dimension to the Tensor, to support images, we can concat all the Tensor in a batch into one tensor and perform the forward operations on that instead of repeating it.

we can concat like this using the concat function introduced in the vectorize branch

    guard let first = data.first else { return  (outputs: [], gradients: Tensor.Gradient(), loss: 0, accuracy: 0) }
    let newData = data.dropFirst()
    let cumulativeData = newData.reduce(first, { $0.concat($1, axis: 1) })