slice_onnx standard/lean leaks

[MirkoCalvi]

Description

lean_slice_onnx() method leaks for a dynamic allocation of the output tensor. The lean version of a mathematical operation MUST NOT dynamically allocate any Tensor!
The standard version of the operation must allocate/create the output tensor tensor, so also computing the output shape, and pass the already created Tensor to the operation_lean() method!

Inside lean_slice_onnx() instead is both computed the output shape (without eaven calling get_slice_output_shape() and the outputTensor.data is dynamically allocated.

All the checks must be inside the standard version.

[huly-for-github]

Connected to Huly®: X_PI-348

[MirkoCalvi]

this is the correct format. get_neg_output_shape() in called so that is also tested in unit_tests