Add's getOutputShape (and possibly others) fails when supplied with two of the same input with shape placeholders

[kostaleonard]

As a machine learning engineer, I want DifferentiableFunction's getOutputShape to succeed when the shape can reasonably be resolved with placeholder dimensions so that I can write complex functions.

Add's getOutputShape (and possibly others) fails in the following case where two of the same Inputs are supplied as arguments. Normally, we wouldn't be able to resolve the output shape, but because it's the same input both times, we should reasonably be able to determine the output shape.

  it should "return its output shape with placeholder dimensions in the same input" in {
    val input = Input[Float]("X", Array(None))
    val addition = Add(input, input)
    val shape = addition.getOutputShape
    assert(shape.isSuccess)
    assert(shape.get sameElements Array(None))
  }