GitLab

new_job -> schedule-task

This is why open-coded sql is a bad idea ...
The Task.abort method does what is needed.

We just show something rather than crashing.

This is an heavy backport wich carries a bit too much baggage ...

Noteworthy:
* we add a db table to store tasks table views per domain (and complete)
* the client checks a hash rather than reloading the whole table: only on changes
  does the reload happen
* a domain selector is introduced
* a dedicated url for task errors is provided
* we provide an extension point `tasks_actions` to allow specification
  of callbacks from unknown tiers

* job_input
* job_results
* relaunch-task

Some js parts are needed for this to work.