Make authorized networks update more resilient

[aptituz]

This change makes the cluster update more resilient by

1. Adding a check for running operations before trying
   to update the cluster. Only afterwards the cluster
   configuration is fetched and updated, to ensure we
   work on the most current state.
2. Adding retries if the cluster update operation fails.
   This is necessary, because two different jobs can
   pass the check for running operations and get to the
   point of trying the update with only a minimal delay,
   and in these cases one of each would fail.

The retry uses an exponential backoff algorithm with
a bit of randomization to avoid many users of this
action becoming synchronized and pestering the GKE
api at once.
Fixes: #75

[aptituz]

In case you would like to see the updated action in action (uh, what a fun to write!):
Here is an example run, also showcasing the problem that is solved with the new action: https://github.com/metro-digital-inner-source/companion-card-service/actions/runs/9479120342/job/26117833835#step:9:1

[cova-fe]

lgtm

[aptituz]

@cova-fe Thanks for the review. I noticed that one of the workflows was failing, as I apparently forgot to run "npm run prepare". I've run it another time and I think you need to approve it again for the workflows to be run again.