Functional.of silently skips layers without inbound layers.

[juliabeliaeva]

Consider the following code:

    val model = Functional.of(
        Input(128, 128, name = "input"),
        Conv2D(name = "conv")
    )
    println(model.layers.size)

This code outputs "1" and the created model has a single input layer, however this was probably not the authors intention. The problem is with the Functional.Companion#topologicalSort method which does not support disconnected graphs and just silently skips layers which are not reachable from the input. Instead an exception should be thrown, or at least an error printed in this case.

[zaleslaw]

Thank you for this issue, agree about the warning or better with exception here