* Delete is_test from activation operators
test=develop

* Revent unneeded changes
test=develop

add approximation for gelu, default value is False (only kernel with eigen is added, remove code for computing gelu with MKLDNN temporarily)

* refine grad maker, test=develop

* refactor tracer stage 1, test=develop

* merge develop to solve conflict third times, test=develop

* add erf op and python interface.

* add fp16 support for erf op.

* add unitests for erf op and its python interface.