Make the system self-recoverable after connection errors

[roll]

• fixes #317
• fixes #172

---

We solve here various issues:

• fix the database connection inside the celery worker after a database restarted the connection (actually this issue #317 is even reproducible locally)
• add 10 minutes soft time limit for all celery tasks - we are ensuring that there is no stale tasks at celery level
• we mark all stale jobs/internal_jobs in the database as errored with a message time limit exceeded if the job is unfinished for more than 10 minutes (we run this cleanup job on SQL errors) - we are ensuring that there is no stale jobs at the database level (e.g. the UI stuck in syncing repos bug)

All these steps applied should guarantee eventual integrity for the system.

[roll]

@amercader Please review

[roll]

@amercader Thanks!

I agree - it should not be a problem while we have only 1 or 2 simple requests there (and anyway inside the already errored session).

[roll]

Probably having regular clean-up jobs could be a better solution. But current should work for now.