Replace (almost) all kernel's GetInput/GetOutput with AllocateTempXXx and DeallocateTempXxx micro context api (#894)

* Replace GetInput/GetOutput with micro context api

* Replace GetInput/GetOutput in kernels with micro_context api. Ther are three kernels (strided_slice, squeeze, transpose)that are not changed yet because:
they put the returned value of GetInput/GetOutput in an OpData structure that is persistent.
It is better to have separate change to opdata of those kernels

* Fix corner case in pad

* Fix cmsis-nn

* Fix xtensa, cmsis_nn, hexagon

* Add virtual destructor.

Some platforms generate warnings/errors that a class that has
virtual functions should have virtual destructor

BUG=observe in internal presubmit

* Fix ReadVariable, AssignVariable and VarHandle to use the micro_context api to get graph

Since PR #819, using GetExecutionPath is no longer the right way to get
graph.  These kernels miss the required change in PR #819.
This PR address this issue.

Additionally, this PR includes some changes to make some internal build
system happy such as avoid casting nullptr to reference

BUG=observe failure in internal example test

* Fix auto merge error manually
Co-authored-by: Nmergify[bot] <37929162+mergify[bot]@users.noreply.github.com>