1. Shrink DoubleBufferReader's buffer size to 3.
2. Add BatchReader an option to discard leftover instances.
3. Fix a MultiPassReader bug on pass count.

Do not use ctor

* Reduce line of codes.
* We can use virtual function for Maker now.
* The implementation does not care what maker holds, it is easier to
refactor later.

* Update tensor_util.h

* Update with moved TensorDesc

* Fix tensur_utils.cu

* Update

* Update

* Update

* Update

* Make tensor_util.cu a symbolic link