Polish fleet API to support cuda collective mode and nccl2 mode

test=develop

* supports distributed classification training
* update API.spec
* fix evenly division in python3
* change "index_range" to "index_num" in shard_index operator
test=document_preview
test=develop

* test=develop, add some comments for Program.clone

* test=develop, add API.spec

* test=develop, refine comments

* refine Program doc and clone doc

* test=develop, refine doc

*  add cuda_is_available
test=develop

* Fix api.spec
test=develop

* fix api doc
test=develop

* fix prepare context redundant code problem, optimize executor by caching create_varaiables
test=develop

* supports collective training in executor

* make fetch_list runable with variables, add more unittest for use_program_cache
test=develop

* fix comment
test=develop

* use unique name for nccl_id

* supports output to stream in program_to_code

* insert sync_comm_stream before regularization; add skip_op_callstack capability in program_to_code

* set op role in collective training

* add collective op role

* remove orig file

* add build optimizer by strategy

* add collective strategy

* refine collective strategy

* add multi-process role maker

* refine strategy building factory so that we can easily plugin more strategy

* scale loss grad in collective sgd transpiler

* add support for distributed fc

* code format

* revert some features for dist fc

* add support for distributed fc training

* fix prepare context redundant code problem, optimize executor by caching create_varaiables
test=develop

* supports collective training in executor

* make fetch_list runable with variables, add more unittest for use_program_cache
test=develop

* use unique name for nccl_id

* supports output to stream in program_to_code

* insert sync_comm_stream before regularization; add skip_op_callstack capability in program_to_code

* set op role in collective training

* add collective op role

* fix comment
test=develop

* remove orig file

* add build optimizer by strategy

* add collective strategy

* refine collective strategy

* add multi-process role maker

* refine strategy building factory so that we can easily plugin more strategy

* scale loss grad in collective sgd transpiler

* add support for distributed fc

* code format

* revert some features for dist fc

* add support for distributed fc training

* test=develop
add collective op unittest standard

* test=develop
remove the test_collective directory

* test=develop
remove the test_collective directory

* remove slicegather test

* code format for reducescatter

* update attr of shard_index_op

* Modify macro nccl_helper

* remove test without distribute

* macro collective_helper

* marcro update

* test=develop
update support python3.5

* test=develop change gpu memory use to 0.1 when test

* test=develop
update ut equal func

* test=develop
set flags to 1.5

* test=develop fix pickle dumple  py35

* test=develop
fix divide in slice and add sync_comm_stream
update atol and rtol to 1e-05
rm shard_index op and test
modify read input from file to read from memory
remove origin_program in framework and add i/o in c_sync_calc_stream

* test=develop update unittest sync operator I/O

* improve the doc of DataFeeder and default_main_program

* update API.spec, test=develop

* fix slice op bug; test=develop

* fix variabel test bug; test=develop

* remove slice while true; test=develop

* Update backward.py:
     - If there is no input grad var in all outputs of previous ops, do not append this op into graph.
     - Only apply this stragety when double backward.
* Update some double backward op.
* Update sum_op to judge whether a tensor is empty by numel or IsInitialized().

* fix code bug
test=develop

* fix get items slice bug; test=develop

* fix unique_name bug; test=develop

* update CPU_NUM config
test=develop

* support avx and noavx core

* add catch and give some log

test=develop

* fix build

test=develop

* add missing package

test=develop

* fix pybind name

test=develop

* fix import error

test=develop

* conbime noavx core

test=develop

* add requirements

test=develop

* fix unkown message

test=develop

* fix api spec

test=develop

* refine and clean

test=develop

* update

* pass dist ut

* follow comments

test=develop

* refine scripts

test=develop

Add Pipeline Concurrency Train Mode:
- Cpp: pipeline_trainer & section_worker
- Python: PipelineOptimizer
- Add a new data_feed type: PrivateInstantDataFeed
- Add a test demo of pipeline trainer and the test model is gnn
- Do not support win32 now

* fix regularizer lod bug; test=develop

* fix exception bug and one_hot expand; test=develop

* add import, test=develop

* fix fill_constant

* fix deconv

* add trainable_statist for bn in dygraph

* support some input tensor remain on cpu; test=develop

* fix input = none; test=develop

* fix unfound bug; test=develop

* fix proto None case; test=develop

* fix bug; test=develop

* fix proto null bug; test=develop

* remove conv check; test=develop

* fix test bug; test=develop

* move fill constant; test=develop

* no change in proto; test=develop

* fix bug; test=develop

* change attr detph name; test=develop

* remove remain cpu; test=develop

* fix bug; test=develop

* merge develop; test=develop

* fix one_hot bug; test=develop

* fix bug; test=develop

* fix bug; test=develop

* fix bug; test=develop

* fix python api bug; test=develop

Note the append_batch_size variable is doing prepend. We should
change the name, but due to backward compatibility, I suggest to
change at v2.0. Not now.

test=develop

* add __str__ method for tensor and lodtensor to support print test=develop

* improve the API Sample of DataFeeder, memory_optimize and release_memory, test=develop

* update API.spec, test=develop, test=document_preview

* tweak the code format of feed API, test=develop

*  update API.spec, test=develop

* improve doc for DataFeeder and default_main_program, test=develop

* test=develop, add gradient sort backward strategy

* test=develop, fix test by add FLAGS_cudnn_deterministic on new tests

* test=develop, test=document_preview, fix all 13 api doc and code

* test=develop, fix rst

* test=develop, refresh API.spec

Fix the following API examples:

paddle.fluid.scope_guard
paddle.fluid.backward.append_backward
paddle.fluid.cpu_places
paddle.fluid.cuda_pinned_places
paddle.fluid.cuda_places
paddle.fluid.in_dygraph_mode
paddle.fluid.CUDAPlace
paddle.fluid.CPUPlace
paddle.fluid.CUDAPinnedPlace

* Fix data and reader related api doc

* Fix data and reader related api doc

Review and fix the example code in some reader related API doc.

These APIs are:

Fix existing API example codes:

paddle.fluid.io.PyReader
paddle.fluid.layers.batch
paddle.fluid.layers.data
paddle.fluid.layers.Preprocessor
paddle.fluid.layers.py_reader
paddle.fluid.program_guard

Add new example codes:

paddle.fluid.io.PyReader.decorate_batch_generator
paddle.fluid.io.PyReader.decorate_sample_generator
paddle.fluid.io.PyReader.decorate_sample_list_generator
paddle.fluid.io.PyReader.reset
paddle.fluid.io.PyReader.start

test=develop

* Add changes to API.spec after changing doc.

test=develop

* Add blanks after python example code

test=develop

* Add blank line at py_reader example code

test=develop

* Merge API.spec

test=develop

* Modify reader.py based on reviewer's comment

test=develop

* Modify API.spec after changing doc

test=develop

* Change reader.py based on reviewer's comment

* Modify example code of decorate_sample_generator

test=develop

* Fix example code of PyReader based on reviewer

test=develop

* fix profiler and name_scope API examples

test=develop

* update API.spec

test=develop

test=develop

test=develop

* Fix bug of gradient interface

* shrink transformer

* Right transformer

* Change from width-first backward to deep-first backward process

test=develop

* Reverse iterator op's input

test=develop

* Polish code

* Change the iteration direction in ingrads' map slots

test=develop

* Polish code

test=develop

test=develop

test=develop