Abstract out user data provider.

[jasonkeene]

It was discussed in #12 that the current method of fetching/caching user data is rigid. It would be nice to have an option, something like user_data_provider, where you can specify your own logic for fetching this data. Use cases we should cover out of the box are:

1. Fetching all user data up front globally for all redactor instances.
2. Fetching all user data up but not cached globally.
3. I'm also thinking of something that would fetch user data based on a URL pattern based on different variables. How we obtain these values I'm not certain. Perhaps have the redactor instance bound to some sort of representation of the current user?