unfortunately some bugs in onnxruntime made the previous approach
crash with certain test topologies, so we're opting for this
instead.

dim restriction: mul must be scalar or 1d
val restriction: mul value must be non-negative

[Test] Changed verification method to a check if all tensor shapes are set and the output of the new node has specific value

switched to giving random names first because the old names can
still contain enumerated names

[Test] Test for shape inference creates now a mixed model and performs shape inference. Afterwards it is executed to check if it still works

[Transformation] Implemented serparate else path for standard nodes in shape inference => All tests now run without errors