Use buffer_info instead of py::array in TensorDataNumpy,
to avoid requesting buffer every time the data() is called.
This is also a workaround for py::array.request() hangup problem.

Merge pull request !5584 from shibeiji/master

Merge pull request !5586 from BowenK/performance

Merge pull request !5601 from panfengfeng/fix_shufflenet_scripts

Merge pull request !5590 from ZPaC/master-fix-server-shard-method

Merge pull request !5585 from YuJianfeng/depend_valid_input

Merge pull request !5545 from 徐安越/master

Merge pull request !5588 from liuzhongkai/activation1_fp16

Merge pull request !5488 from zhangbuxue/add_check_tensor_before_compile_graph

Merge pull request !5593 from liuwenhao/master

Merge pull request !5553 from zhouyaqiang0/master

Merge pull request !5314 from liuxiao93/implicit-conversion-api

Merge pull request !5574 from lihongkang/lhk_master

Merge pull request !5583 from sunsuodong/fix_concat_slice

Merge pull request !5578 from ZPaC/master-fix-sparse-slice-leak

Merge pull request !5568 from yoonlee666/master

Merge pull request !5559 from lyvette/tflite_parser

Merge pull request !5566 from mengchunyang/master

Merge pull request !5479 from lyvette/master

Merge pull request !5551 from ZPaC/master-fix-stuck-bug

modify log

Merge pull request !5558 from wangmin0104/master

Merge pull request !5563 from yeyunpeng2020/master

Merge pull request !5564 from cjh9368/fix_static_error

Merge pull request !5517 from limingqi107/master

Merge pull request !5202 from ling/sr

Merge pull request !5513 from cjh9368/aware_quant

Merge pull request !5552 from lyvette/lite

Merge pull request !5547 from mengchunyang/master

Merge pull request !5496 from jiangzhiwen/bug/csv_char_to_int

Merge pull request !5556 from yangruoqi713/lite

Merge pull request !5554 from zhanyuan/master

Merge pull request !5522 from liuzhongkai/fp16