Optimize amp for multi-gpu to enable FP16 gradients transfer across gpus

Fix float16 optimizer

init new amp, optimize inserting cast op for batchnorm

* remove unused inference_transpiler unit-tests

test=develop

* remove InferenceTranspiler usage in quantize_transpiler.py

test=develop

* add multi process reader
test=develop

* Cherry-pick fix random Python3 CI failure.

In some tests, SWEs used "print('xxx').format('xxx')". The syntax
is only supported in Python2, not python3. However, since those
lines are related to data download, if the CI machines already have
the data, it passes CI tests. That causes random failure.

* Cherry-pick: disable CUDNN case of test_warpctc_op

Also temporary disable a unit test. The test will be fixed under high priority.

* add data parallel batch

* Add LAMB optimizer

* Expose LAMB Optimizer's APIs

test=develop, test=document_preview

* Cleanup code & doc

test=develop, test=document_preview

* Update lamb optimizer's formula

test=develop

* init auto loss scaling

test=develop

* change API.spec

* change ifelse to switch and use reduce_sum to optimize checking isfinite

test=develop

* Remove redundant code

test=develop

Update the folder name generation mechanism for saving the quantized model and weights.
The folder name would be unique by adding the timestamp postfix.

test=develop

* Init mixed precision training interface

* Add fp16 test script

test=develop

* All initializers support float16

test=develop

* Code cleanup & add more code annotations

test=develop

* Update API spec

test=develop

* Add usage example in doc

test=develop

* Split test_calibration test to two tests

test=develop

* Modify CMakeLists file
test=develop

test=develop

test=develop

* Add DecoupledWeightDecay

* fix cdn issue, test=develop

* fix cdn issue, test=develop

test=develop

1.Remove the code for setting mkldnn environment in the test_calibration.py;
2.Update the cmake file for MKLDNN environment enabling;
3.Update the INT8 inference doc.

test=develop

* Add INT8 calibration support in Paddle package; test=develop

* Enable performance measurement in INT8 calibration unit test

* Enable mobilenet UT in separate test class; use download cache by paddle download utility and cache unzip; and fix typo; test=develop

* Extract cache_unzipping function for reuse; format code style; test=develop

* Simplify the test code by define a combined function for both downloading and unzipping; test=develop

* Add the calibration tool code for int8 inference and focus test.

* Fix the calibration tool per the review comments.

test=develop

* Update the calibrator doc and remove extra line.

* Fix the invalid is_negative_input attr set on Mobilenet.

* Add the comments and fix the format issue.

test=develop

* Update the CMakelist.txt for Calibration PR.Disable the Calibration UT if not enable MKLDNN.

test=develop

* Update the CMakeList.txt.

test=develop

* Disable the test_calibration case on WIN and MAC.

test=develop

* Add the missing brackets.

test=develop

* Remove the outdated map operator which not supported on Python3.

test=develop

* Fix the style issue.

test=develop

* 1.Update the CMakeList.txt to disable calibration tool ut when the WITH_MKL is not set;
2.Add the workaround to enable the FLAGS_use_mkldnn for PR_CI(PADDLE).

test=develop

* Fix the typo and format the License header.

test=develop

* 1.Add and Update TODOs per review comments.
2.Code clean.

test=develop