[cherry-pick] [NPU] apply npu cache offline model to other devices, test=develop (#3925), test=release/v2.6 (#3933)

* [NPU] enhance cache offline model, test=develop

[cherry-pick][OPENCL] Fix opencl create cmd with prop, fix int16 model for fc opencl kernel. (#3918)


* [OPENCL] Fix opencl fc int16 model bug caused by fc kernel (#3900)

* fix opencl fc kernel caused int16 model weight abnormal. test=develop

* fix opencl fc kernel caused int16 model weight abnormal. test=develop

* [arm]add 2x2s2p1 pooling  (#3705)

* fix pooling bug and speed

* add 2x2s2p1 pooling. test=develop

* fix conflict, test=develop

* fix conflict in wino

* [arm] add 3x3s1 Winograd int8 (#3767)

* fix: winograd support unsame pad
test=develop

* feat: add winograd int8 kernel
test=develop

* fix: style fix
test=develo

* fix winograd_int8 ut sgement default. test=develop

* close basic_test, test=develop
Co-authored-by: NMyPandaShaoxiang <txg4794@163.com>

* fix xiaodu crash in gemm prepacked

* in huwen phone, 3x3s2p0 avg pooling will rand crash, other phone does not have this feature

* [arm] update con int8 kernel choose (#3834)

* fix conv int8 kernel choose and sooftmax compute bug

* change axis_size = 4 kernel choose, test=develop

* fix format. test=develop

* fix format.test=develop

* fix build test=develop

* fix buiild error test=develop

* fix wino_int8 computte erroor. test=develop

* Update the link to debug, test=develop, test=document_fix (#3870) (#3871)
Co-authored-by: NMyPandaShaoxiang <txg4794@163.com>
Co-authored-by: Ncc <52520497+juncaipeng@users.noreply.github.com>

* avoid saving shared params repeatedly (#3561)

* test=develop
Co-authored-by: Nzhupengyang <zhu_py@qq.com>
Co-authored-by: Nhuzhiqiang <912790387@qq.com>

* move train demo code, test=develop

* test=develop

[Cherry-pick][BUG FIX] Fix the issue that Android static library can not be linked properly  (#3695)

fix the demo of opt saved can't work on windows

 fix the doc

[cherry-pick] [BUG FIX] Fix the issue that paddle-lite python lib can not inference on mac env (#3684)

* Update benchmark doc for v2.6, test=develop, test=document_fix (#3660)
* [Doc] Update shell for make benchmark_bin (#3661)

* test=develop, fix the x86 demo can't work on windows

* test=develop,test=document_fix, add the doc of python demo

* test=develop, add the document of windows compiled

* test=develop, modify the x86 demo on windows

* test=develop, modify the name of windows batch file

* test=develop, fix the doc

[Cherry-pick] [Compile] [fluid-lite-subgraph] Fix fluid-lite compile error, because of not defined LITE_WITH_LOG (#3611)

* update cuda demo.

* [LITE] Set high perf mode as default in mobile_light deme; Fix Precision Profiler caused by X86 header; Enhance profiler for opencl. test=develop
Co-authored-by: NHappyAngel <chenjiaobuaa@126.com>

* v2.6 supports mask_demo, test=develop

* update cuda compile. test=develop test=release/v2.6.0