Since the earliest days of edX, we've run gunicorn with preload_app as
True. This was primarily because the first handful of courses were using
the XMLModuleStore, which was extremely slow to initialize because it
had to read entire courses from disk and load them into memory at
startup.

XMLModuleStore courses stopped being created with the introduction of
Studio, and we've long since removed them from edx.org. Note: This
should not be confused with importing courses that are authored in XML.
XMLModuleStore courses are courses who use the disk instead of Mongo
for content persistence, and require a server restart for content edits
to show.

This commit sets things back to how Django is more commonly deployed
with gunicorn. It will make server startup time somewhat more expensive
because each worker process does init work, but it should make it
easier to safely enable features like database connection pooling
without having to worry about shared resources.

[PERF-305]

Added `EDXAPP_LMS_ENV_EXTRAS` and `EDXAPP_CMS_ENV_EXTRAS` to default vars in order to be able to have different values for the `env_extras` between LMS and CMS.

By default supervisor will only automatically restart services when they
exit code is not 0 or 2.
Edxapp workers sometimes exit with a code of 0 on certain network
failures, so make sure they are restarted even if exit code is 0.

We had a persistent problem of workers getting stuck when trying to stop
during 12.04 native installation.  This fixes the problem, while still
allowing active tasks to complete.

This reverts pull request #2133, and removes the stopasgroup setting
that was causing the trouble.

pre_supervisor expects to run the migrate tests as the supervisor user,
which does not have sudo.  Add an env variable flag that allows you to
claim "I know what I'm doing, don't sudo".

This allows us to run `edxapp-migrate-lms --list` for migration checking

When we have multiple DBs, this will do the right thing.

Django 1.8 removed the syncdb command so we can no longer run it, don't
imply that we are running it in documentation/messages, also don't
provide the scripts used on devstack/sandboxes to run it since they'll
never work anyway.

Additionally - remove yet another openid workaround which is unnecesary
since the Django 1.8 upgrade.

In particular LocMemCache doesn't have this option.

For more details, see:

http://stackoverflow.com/questions/23358787/memcache-get-returns-wrong-object-celery-django

According to the documentation TERM should do a graceful shutdown:
http://celery.readthedocs.org/en/latest/userguide/workers.html#process-signals

but this is not actually the case:
https://groups.google.com/forum/#!topic/celery-users/t8g5KvIvQZ8

SIGINT on the other hand seems to do what we want.

COMMON_ENABLE_NEWRELIC turns on server monitoring and is on for all
sandboxes.  App monitoring should not be enabled for all sandboxes and
since edxapp workers is on all sandboxes, surprise newrelic!

Update default tasks per child for high mem queue.

Part of the file name template is eaten up by Python's `%` formatting,
resulting in log files missing the `s` of "stderr" and "stdout".

This adds some helper scripts for managing the edx server.

Writes out yaml files in addition to the json files
to a new directory in the (by default) /edx/etc/edxapp
directory