fix/avoid locks being unrecoverable

[aspedrosa]

If we performed a deployment upgrade while some celery tasks were executing, some redis locks would be unrecoverable.
With this PR:

1. created a separate redis db for locks.
2. when the dashboard worker starts:
   1. clears all locks
   2. starts tasks for all pending uploads that might have been interrupted. Since those execute operations over the database on a transaction, we can just restart the upload process because the previous progress was discarded.

Some other improvements that could be implemented: if there are several uploads pending, to the same database, only resume the most recent one and cancel the others.

[joaorafaelalmeida]

While we do not approve this PR, we can solve this issue using the following:

1. docker-compose restart dashboard_worker
2. docker-compose logs dashboard_worker if any task restarted
3. access redis container
4. access redis database redis-cli -n 0
5. Execute keys *. It should display something like "lock: lock...", "workers_updating..."
6. Finally, execute flushdb

[joselfrias]

Reviewed.