Fix munin zeros

[noirbizarre]

This PR ensures that the udata worker status --munin always output tasks count even if count is zero.
Otherwise Munin sees it as "no data" (ie. vertical splitter in graph) and display count as -nan instead of 0. It was also making Munin diagnosis difficult when a queue is empty.

This also ensure only tasks for monitored queue are listed in udata worker status --munin-config reducing the number of tasks by graph (extras tasks were always displayed as -nan. As a side-effect, caching tasks is not required anymore as it does not hit all the Celery workers anymore to list them.

A udata worker tasks command has been added to display all tasks and their queue/priority.

Some test jobs have been renamed to ensure they all have the same test- prefix and they are filtered out of munin commands.

The documentation link to the munin plugin has been updated.

[abulte]

it does not hit all the Celery workers anymore to list them.

I'm not sure I understand why 🤔 and thus why we don't need caching anymore.

[noirbizarre]

As stated here, we were using cache because we were using Celery inspect() which is querying the state of every worker (the w here).
It was working with the following limitations:

• the more we have workers, the more inspect() takes time (ie. query duration * number of workers)
• it only works when workers are started and for every queues
• it only works if the works if the workers are unfiltered (ie. when using one of --queues/-Q, --exclude-queues/-X, --include/-I)
• it only works if workers are reachables

If has been replaced by celery.tasks which doesn't query anything, these are only the tasks registered to celery in the code and so it doesn't depend on the workers at all (no network roundtrip, not state based)

But it's how I understand this comment. If cache is still required, I can put it back.

[abulte]

@noirbizarre Ok thanks! Looks good.