It will be used for LoD information in LoDTensor since LoD is a copy
on write field.

It is pretty slow for copying LoD information between operators. For
resnet it will cost roughly 10% time of whole time, including reading
data.

* send_recv variable

* delete unused logs

* fix ci failed

* update

* resize tensor before tensor copy

* add selectedrows unit test

* check rows

* "remove cudnndevicecontext"

* "remove unused init code"

* "fix hash functions"

* "fix typo"

* "fix based on comments"

* "follow gogle style"

* "fix based on comemnts"

* "fix frigled test gradient of rnn"

* "fix based on comments"

* fix bug in SetAttrDescVisitor

* add comments

* "fix data transform"

* "data transformer"

* "add device pool"

* "add test"

* "fix ci"

* "fix datalayout implementation "

* "fix based on comment"

* add helper function to get appropriate DeviceContext