Execute `failure_detector` at each gossip round

[Gau-thier]

Hum, this is wrong, but works regarding what we do.
On the SWIM paper, I think the failure_detection is executed at each round of the gossip. Since we do that before each round, this is kinda ok. But I think we do need to add two things:

• round ID in each failure_detection message
• failure_detector_config as argument here, because user must be able to set whatever is relevant for his/her business case.

Originally posted by @gsebil08 in #33 (comment)