layer run-context should perform caller checks

[kparichay]

layer run-context should perform caller checks before providing access to tensors. For example, accessing a tensor requested with FORWARD_FUNC_LIFESPAN in calcDerivative() must throw.

[taos-ci]

:octocat: cibot: Thank you for posting issue #1592. The person in charge will reply soon.

[kparichay]

We can use __builtin_FUNCTION() as a parameter to the getters to filter out the functions which can be provided access to the requested tensor.

[kparichay]

Also, add checks for accessing the tensors based on the mode of execution - training / inference / etc.