use jinja 2.11.3 in python lower than 3.6.2

* move trace into api.yaml

* add trace unittest

* fix trace test

* fix generate op

* add codegen for get_expected_kernel, add argument mapping for selected_rows kernels, fix other bugs in codegen for operators.
* move bernoulli, erf, mv, poisson, trunc, erf to api.yaml and corresponding backward api to backward.yaml
* generate EmptyGradOpMaker for ops without grad op
* add code to generate all possible kernel signatures for infrt

Add Code Generation for operators,  op makers and argument mapping functions (#41772)