bake: Enable multiprocess baking. Baking is now done by running a worker per CPU, and sending jobs to them. This changes several things across the codebase: * Ability to not cache things related to pages other than the 'main' page (i.e. the page at the bottom of the execution stack). * Decouple the baking process from the bake records, so only the main process keeps track (and modifies) the bake record. * Remove the need for 'batch page getters' and loading a page directly from the page factories. There are various smaller changes too included here, including support for scope performance timers that are saved with the bake record and can be printed out to the console. Yes I got carried away. For testing, the in-memory 'mock' file-system doesn't work anymore, since we're spawning processes, so this is replaced by a 'tmpfs' file-system which is saved in temporary files on disk and deleted after tests have run.

--- a/piecrust/sources/array.py	Sat May 30 15:41:52 2015 -0700
+++ b/piecrust/sources/array.py	Fri Jun 12 17:09:19 2015 -0700
@@ -1,5 +1,6 @@
 from piecrust.sources.base import PageSource
 from piecrust.sources.mixins import SimplePaginationSourceMixin
+from piecrust.sources.pageref import PageRef
 
 
 class CachedPageFactory(object):
@@ -30,7 +31,7 @@
 
 class ArraySource(PageSource, SimplePaginationSourceMixin):
     def __init__(self, app, inner_source, name='array', config=None):
-        super(ArraySource, self).__init__(app, name, config or {})
+        super(ArraySource, self).__init__(app, name, config)
         self.inner_source = inner_source
 
     @property
@@ -41,3 +42,6 @@
         for p in self.inner_source:
             yield CachedPageFactory(p)
 
+    def getTaxonomyPageRef(self, tax_name):
+        return None
+