Stale config is never cleaned up on ENI detach, shadowing the new config when the interface is renamed on re-attach

[joeysk2012]

When an ENI is detached, the udev rule in udev/99-vpc-policy-routes.rules runs
systemctl disable --now refresh-policy-routes@$name.timer policy-routes@$name.service,
but nothing ever deletes the generated configuration. policy-routes@.service defines
only ExecStart, so stopping the unit executes nothing, and the remove) branch in
bin/setup-policy-routes.sh — the only code that removes 70-<iface>.network and
70-<iface>.network.d/ — has no caller anywhere in the tree (its sibling
stop|cleanup) is an explicit no-op), making it effectively dead code.

The orphaned config therefore survives the detach with ec2net_alias.conf still listing the
pre-detach secondary IPs, and because create_if_overrides writes
[Match] MACAddress=${ether} rather than matching on interface name, it continues to
match the same ENI. If that ENI is later re-attached and the kernel assigns a different
name (e.g. enp40s0 → enp41s0), two .network files in /run/systemd/network match
the same MAC and systemd-networkd applies the lexicographically first — the stale one.
From that point refresh-policy-routes still recomputes the alias list correctly, but
writes it to 70-<newname>.network.d/ec2net_alias.conf, a drop-in belonging to an
inactive .network file, so removed secondary IPs are never withdrawn from the OS and
neither networkctl reload nor a full systemctl restart systemd-networkd recovers.