add paddle.tensor.linalg.diag API, diag_v2 OP and CUDA kernel.

* Support LoDTensorArray in reverse_op test=develop

* polish en doc and unittest code test=develop

* refine sample code test=develop

* add example of LoDTensorArray test=develop

* fix typo test=develop

* fix error log: resahpe, range, reverse.

python API(get_tensor_from_selected_rows and tensor_array_to_tensor)error message enhance, test=develop (#23636)

* error message enhancement create_parameter create_global_var, test=develop

* use six.PY2 to use long, test=develop

* fix input dtype for create_parameter, test=develop

raw doc: bool, float15, float32, float64, int8, int32, int64, uint8.
new doc: bool, float16, float32, float64, int8, int32, int64, uint8. 

* Add the support of bool list for assign value, test=develop
* Fix the assign op test case for bool dtype, test=develop

API(has_inf, has_nan, cumsum, fsp_matrix, increment) error message enhancement, test=develop (#23769)

* API(argsort, argmax, argmin, cast, diag) add error message，and the same time add the test cases for those apis.

Those ops include，scale， sum， sums，unique_with_counts，unique，
wherre， add error message and test case

Add the error raise for those cases
aassign isfinite linspace ones_like zeros_like zeros ones

* API(argsort, argmax, argmin, cast, diag) error message enhancement, test=develop

* refine example code in en doc, test=develop, test=document_fix

* update shape to make it meaningful, test=develop, test=document_fix

* API (Switch) error message enhancement. 

* fix bug: dtype of out in api isfinite is set incorrectly. The dtype should be bool. 

* add support for value tensor support of fill_constant Op

* Add unittest for transformer prediction in dygraph_to_static.

* fix bug in fill_constant api. 

* Make transpose support size 0. test=develop