Discovered that AJAX requests to the API from the web view were
using certificates in people's browsers even if they were logged
out of the Django session.  So when they wanted to view the page
as a public user they still saw some extra information.

No security risk since they needed to have a valid certificate
anyway, but a bit annoying.

We've had some bad cases where a log entry is created, but the
file is not saved due to filesystem errors. So we put the file
creation in a try-except block and if it fails, delete the log
message that was just created before re-raising the exception.

Previous arrangement was not including zero-valued fields. We now
include all non-None fields.

Due to the fact that we are using multiple attribute authorities,
there are some cases (specifically email) where there are multiple
values provides for an attribute, separated by a semi-colon.

This update handles those instances and takes the first value in
the list.

Use the 'sync' worker still due to issues with handling file
uploads when gunicorn uses at least 6 but possibly more than 2
workers.

Suggested keepalive time is 1-5 seconds for standard deployment,
but it should be much higher for deployments behind a load
balancer. We are hoping this might resolve the SSLEOFErrors which
have been seen on the AWS deployment.  Traefik's keepalive time
is hard-coded to 90 seconds so we want gunicorn's to be longer than
that, so we set it to 100 s.

Increase number of attempted restarts for LVAlert Overseer in
a container from 3 to 100. With only three attempts, it doesn't
even consistently get going again without going into a FATAL state
when the openfire server is simply restarted.

These settings were previously handled in settings/container/production.py,
but we want to do some testing and I don't see any issues with having
them in the base container settings (settings/container/base.py).

Upgrade gunicorn to 19.9.0 and switch to an asynchronous
worker class

Instead of looping over EmailMessage.send() calls, we now get the
email backend and send a list of messages. Note that the backend
sends the messages individually still but this cuts out some of the
overhead.

Prevents server errors from being generated when the filename is
"bad" and when the user specifies a non-integer version. This is
for superevent files; event files are handled separately, although
this should be unified at some point.

We use get_or_create now to add labels to a superevent in order
to prevent IntegrityErrors when a user makes the same request
too rapidly and both sneak by the initial check.

overseer seems to always exit with an exit code of 0, which means
that 'autorestart=unexpected' does not restart it. So we set it
to 'true' instead.

Not feasible to use multiple DBs for read or write operations at
present. We're going to have to revisit this at some point. We leave
most of the code in place but commented out.

Add capability for passing kwargs to AlertIssuer classes'
issue_alerts() method.  We take advantage of this in the update
alerts for checking for FAR changes and NSCAND state changes.