GitLab

This way, the envelope id is known and we avoid errors when getting a currently
empty envelope

This will make forensic easier

This should fix zombies processes (RUNNING with no jobs)

In some case, null fragments are received before the first actual fragments.
The envelope storage should just ignore them and not fail.

- added commands 'info' and 'stack'

- sort nodes and edges to avoid finding false differences
- always save new versions, even if identical to previous version

A vicious loop variable reference was anihilating the careful filtering of matching pipelines.

Add a new configuration entry: http.webhook.payload-type, that determines how to post
the data to the target url. The possible values are:

- 'full-request' (default): same behavior as before, the full request is serialized and posted
- 'envelope': post only the envelope and expects no content in return (ideally a 204 status).
- 'message-in-request': post individual messages in a complete request. The envelope id is
  not changed.
- 'message-in-envelope': post individual messages in an envelope
- 'message-content': post the messages contents, one by one

The webhook.url setting can now contain a '{msgtype}' placeholder that will be replaced
with the message type (only if payload-type is 'message-*')

In the 'full-request' mode, the big envelopes should now be posted complete, not only
the header.

--HG--
branch : 3.2.x

--HG--
branch : 3.2.x

If actual collection stops, the time since lastCollect keep growing, which caused
the wait time to be reduced to the minimum (1ms), hence a constant update of
the metrics, leading to a heavy load on the db.

--HG--
branch : 3.2.x