Fix the HierarchicalSigmoidGradOpKernel and refine the codes. Now hsigmoid_op is same with V2 implementation and can pass gradient check.

* change im2sequence for ctc batch inference

* Update im2sequence_op.cc

* change im2sequence for ctc batch inference

* update

* change PR by comment

* fix ocr test error

* fix test_im2sequence

* modify the old name to standard name

* fix test_layers failed

* move to platform

* "move init from framework to platform"

* "remove used init"

* "fix ci"

* "fix ci"

* "fix generic"

* "fix ci"

* "fix ci"

* "fix ci"

* "disable fragile test"

* Make the CUDA kernel of concat correct and fix unit tests.

Add an interface to set the number of threads for math function, and set the default value to 1 for inference.

* Fix a bug related to dispensable inputs and refine the inference unittest.

* Fix the use of dispensable inputs in reshape_op.

* Polish the enforce statements.

* Fix an English writing typo.

Rename MatDim to MatDescriptor

* adding operator to blas deps list

* use device_context instead to solve cycle deps

* Fix CPPLint issyes in gru_kernel.h

* Fix CPPLint issyes in gru_kernel.h

* Fix Compile error

* fix __shfl_down_sync_ of cross_entropy

* use reduceSum

* "fix ci"

* Fix cpplint issues in sequence_padding

* Fix typo in cu file

* Fix dependencies of sequence_padding

* Add include