* Clear no persistable tensor array before predicting, fix crash when predicting with gpu debugging mode

* Fix code style

* Add fetch API for java, refine android log

* add prepare data and model script, modified corresponding docs:development_android and development_android_GPU

* Add verified multi-threading supported gemv_trans_mx1

* add comments for alloc and free

* optimize reduction progress of gemv_trans_mx1

* add macro for gemv_trans_mx1

* Fix multi-threadings incorrectnesss of armv7 neon gemv_trans_mx1

* add test_gemm_accuracy

Clear no persistable tensor array before predicting, fix crash when predicting with gpu debugging mode (#1548)

* Clear no persistable tensor array before predicting, fix crash when predicting with gpu debugging mode

* Fix code style

modify FPGA doc close #1544

improve program optimize, add assign value and flatten2 op

Replace C++ typeid with self-defined type_id to avoid compiler difference, and fix gpu bug

anchor_generator support marker for FPGA track #1537

clean interrupt signal at the beginning of bypass in FPGA track fixed#1531

Fix multi-threads sgemm and conv/add/bn/relu fusion bugs

update EW config in api.cpp in FPGA track fixed#1526

add new interface function for FPGA track close #1479

fix the dealign bug in fetch op FPGA track fixed#1521

fix memory leakage problem close #1523

fix a bug in sigmoid kernel in FPGA track fixed#1518