When querying Server .../depsgraph, we just take the Last-Modified header
value as string, and store that without doing any parsing. This prevents
issues with timezones getting lost, formatting going wrong, etc.

Still rough & untested.

Also removed superfluous \n in log messages.

By default logs only WARNING and higher. I've added -debug and -verbose
flags for respectively DEBUG and INFO level logging.

Most logger calls are still just INFO-level, will be migrated in a future
commit.

A closed timer channel now no longer results in an infinite loop. Hopefully
this resolves the hanging shutdown.

N defaults to 60 seconds, which is fairly short. The flamenco-manager.yaml
file can be used to increase this timeout.

We no longer use PATCH on tasks to change their status; only the task
batch-update mechanism is used.

The returned tasks have been set to 'cancel-requested' on the Server, and
should be canceled on the Manager & Workers as well. This is implemented
by locally setting the task status to 'cancel-requested', which is seen
as not-runnable by workers, which automatically results in a cancel.

Also simplified some code; after sending a task to a worker, the Manager
no longer sends a PATCH command to Server to change the task status. The
worker is responsible for setting the task status to "active", since those
status changes are nicely batched up and retried if necessary.

This includes checking for any server-side changes. If there are, the
queued task is updated accordingly, and the queue is re-examined.

Connection errors, redirects and Server-side internal server errors are
ignored, and the task is considered "unchanged" in those cases. Errors
in the 400-range (Not Found, Forbidden etc.) result in the task being
marked as "canceled" in the queue. The task is NOT removed from the
queue, so that the information about the task being unavailable to workers
is explicitly available.

Furthermore, this commit introduces gocheck for unit testing, which gives
us setup and teardown functions.