Use Flask.before_request instead of Flask.before_first_request to work around the re-entry deadlock introduced in Flask 0.11

[joeyespo]

Fixes #182.

Flask changed its behavior in 0.11 by making before_first_request deadlock if the handler causes another request before finishing. The _before_request hook is still re-entrant, so this can be fixed without changing much code.

Note that Flask's change in behavior addresses some real problems. So a larger, more proper fix later on could be to change Grip._retrieve_styles (or more specifically, Grip._get_styles) to access the cached files directly instead of using a request to do so.

[joeyespo]

Note: The test failure is coming from pypy3 (which currently implements the now-outdated Python 3.2 version). Ignoring this since later versions of Python 3 are more stable, and the next version of Flask works around this issue. See this Flask issue for details.