Created by: jacquesqiao

• 支持 l2 decay https://github.com/PaddlePaddle/Paddle/pull/15100
• 优化 assign https://github.com/PaddlePaddle/Paddle/pull/15080
• 多线程adam加速 https://github.com/PaddlePaddle/Paddle/pull/14890
• nce and hs support prefetch https://github.com/PaddlePaddle/Paddle/pull/14893
• 修复adam胡乱新建临时变量的问题(cpu下) https://github.com/PaddlePaddle/Paddle/pull/15111
• 修复分布式模型save/load/checkpoint问题 https://github.com/PaddlePaddle/Paddle/pull/14854
• 加上remote_prefetch参数 https://github.com/PaddlePaddle/Paddle/pull/15316
• 修复nce下按权重采样会有-1出现的问题 https://github.com/PaddlePaddle/Paddle/pull/15463
• 多线程adam运行速度会变慢 @seiriosPlus