internal: Get rid of the whole "sub cache" business. * Compute cache keys up front, so the cache directory is only chosen once. * Buffer up config variants to apply before loading the config. Makes it possible to cache variant-resulting configs, too. * Make a factory class to reuse the logic that creates the `PieCrust` object correctly for multi-process workers and such. * Add a test.

import os
from invoke import task, run


@task
def genmessages():
    root_dir = 'garcon/messages'
    out_dir = 'piecrust/resources/messages'
    run('python chef.py --root %s bake -o %s' % (root_dir, out_dir))
    os.unlink('piecrust/resources/messages/index.html')