core: Start of the big refactor for PieCrust 3.0. * Everything is a `ContentSource`, including assets directories. * Most content sources are subclasses of the base file-system source. * A source is processed by a "pipeline", and there are 2 built-in pipelines, one for assets and one for pages. The asset pipeline is vaguely functional, but the page pipeline is completely broken right now. * Rewrite the baking process as just running appropriate pipelines on each content item. This should allow for better parallelization.

--- a/piecrust/cache.py	Sat Apr 29 21:42:22 2017 -0700
+++ b/piecrust/cache.py	Wed May 17 00:11:48 2017 -0700
@@ -5,7 +5,6 @@
 import codecs
 import hashlib
 import logging
-import collections
 import repoze.lru