Created by: backyes

The PR adds some demos and DOCs for local and distributed sparse training.

1. add QuickStart for sparse training
2. add Performance Analysis on Sparse Training
3. add Internals for Sparse training
4. add some FAQ
5. Fix tiny sparse related error in dataprovider DOC.

related issues and PRs: https://github.com/baidu/Paddle/pull/144 https://github.com/baidu/Paddle/issues/137 https://github.com/baidu/Paddle/pull/172

previous comments: https://github.com/baidu/Paddle/pull/144/files/b7b9b39ae631c9daef2484e6f8b1cbe39501ac49#diff-dad8ccff15b7bf8f5ee59298f0345835.