refactor and optimize document parsing

[liuq19]

To fix #34

Tasks:

• support parse owned dom in Rust struct

• use thread-local preallocated buffer

• add more apis as serde_json: such as from and from_iter and so on.

Design:

1. use memory poll allocator and smallest node layout
2. shared parts support sync, such as allocator
3. no explicit allocator args