macOS fleet self-heals from Unknown nodes and wedged xcresult processors

Fixes an incident where macOS CI fleet nodes could get stranded in a Ready=Unknown state and xcresult processor VMs could silently stop consuming jobs for hours while still appearing healthy. Recovery now happens automatically: the kubelet self-heals nodes out of Unknown after a heartbeat gap, the processor VM boot chain times out and retries instead of hanging on a tailscale control-plane hiccup, and the host’s en0 NAT that lets VMs reach the public internet is re-asserted every minute so a churn-induced flush re-converges within the minute. Net result for users: hung or unavailable Scaleway Apple Silicon runners recover on their own rather than needing a manual recycle.