Investigate tunnel registration GC spike w.r.t KAS shutdown

We have some spikes in the garbage collection metric for the tunnel registry:

image

These seem to loosely correlate with deployments (the green thick blocks on the X-axis) and especially with the delayed KAS shutdown:

image

It's not happening for all deployments but those certainly relate.

This is an indication for some bug when handling KAS shutdowns and tunnel unregistration.

So far, this has no impact on KAS availability or tunnel routing performance.