As a convenience, the generateed `remotecontentimport` task was setting
`node_ids` to an empty list if it wasn't specified. However, an empty
list and `None` are 2 different cases. An empty list is used when no
node IDs are to be imported, and `None` is used when all node IDs are to
be imported. Even though we're not actually downloading all nodes here,
fix the semantics and require that the thumbnails tasks explicitly
request no nodes. This will be used later in testing.

This will be used in testing to check that a task was actually enqueued.

The storage hook depends on the `BackgroundTask` containing the current
job ID. However, it's possible the job will change state and the hook
will run before the job ID has been saved. That would prevent the hook
from properly synchronizing the updated job state. To prevent that, lock
the database while the task is being enqueued until the job ID is saved.

By default set the log level to INFO so there's useful information on
failures. Task handling is multithreaded, so add the thread name to help
debugging.

The logs won't be shown if the tests pass, but if they don't the
messages can be invaluable for debugging failures.

Learning Equality obviously has no use for the test suite constantly
telling it about an ephemeral deployment. More importantly, the ping
task can hang and interfere with our own tasks being scheduled.

In order to exercise the collection downloader, we need channel data.
The `create_channeldb_json.py` script and `db.json.template` file are
used to create JSON files representing fake channels. The generated JSON
files contain the data necessary to create sqlite channel databases with
sqlalchemy as well as the content data inlined. 18 fake channels have
been created and will be used in later commits.

The `create_contentdir` function takes the fake channel DB JSON files
and creates a content directory with the channel databases and content
files for testing. That's made available as the `contentdir` pytest
fixture.

A standalone script is provided as a convenience and for exercising the
functionality outside of the test suite.

This runs an HTTP server for the test content so that we can import
channels and content during tests just like they were being imported
from studio.

Provide test collections using all the packs in the current
endless-key-collections release. The fake packs have very regular
structures and use the fake channel data.

These exercise all of the API endpoints that don't interact with the
download manager.

This adds a few pytest fixtures for running channel and content import
jobs. Beyond creating a facility and a facility user with appropriate
permissions, 2 fixtures for handling interactions between Django and
SQLAlchemy. Some of this is lifted from Kolibri and/or depends on
Kolibri internals. We'll see how well they hold up over time.

While here, add a few log messages to aid when debugging failures.

It seems that Kolibri's task worker can hang occasionally, so use
pytest-rerunfailures to mark tests that can be retried on failure.

This is a basic smoketest that the collection downloader can be run with
any collection specified. Unfortunately, Kolibri's task worker seems to
hang sometimes during either the foreground or background downloading.
Use pytest-rerunfailures's `flaky` mark to try again when that happens.

Fixes: #778