Problem:
Currently, the "data" tokens don't have the information about if the "data" is inside the raw/endraw block, or not. This limit the Extension object's ability to treat the "data" tokens differently. I'm developing a jinja extension which is limited by this.

Feature:

1. Allow Lexer to optionally choose to emit the raw_begin and raw_end tokens, v.s. ignore them (current behaviors)
2. Allow Environment to switch the lexer provider, so that you can inject your customized Lexer instances. For example, a customized one when an Extension is added, or when it's in testing.