Performance optimization

[zhaozg]

As describe in https://github.com/luvit/luv/pull/331, There is a great possibility to get better performance.

For example, from the following aspects:

• Using luaL_ref to store the lua_State in the registry/avoid using a string key
• Store an internal map of lua_State* -> int ref pairings.
• ...

Long-term task.