What currently happens is that in most cases people come to our team asking for help having only must-gather. Whenever host networking issues are suspected, the first thing to look at is output of `ip a` and `ip r`. Sadly, neither is available via must-gather. As a result, we play ping-pong because we request sosreport which often proves impossible to collect (e.g. when the input was a failed Prow job).

To immediately fix this issue, we want to have `ip a` and `ip r` collected via must-gather.

Also, to make Prow more robust, we want to collect sosreport from all the nodes whenever a job fails.