Dangling gradient should be always zero

[zhoonit]

As we are having output that is being dangled, we have to manually set the incoming derivative to zero for this case.

Later we can have a 'constant' (RV of a tensor) which we should integrate this to the tensor management scheme.

[taos-ci]

:octocat: cibot: Thank you for posting issue #1804. The person in charge will reply soon.