client/doublezerod: set tunnel interface down before deleting

[snormore]

Summary

• Set tunnel interface administratively down (LinkSetDown) before deleting it during teardown across all service types (IBRL, edge filtering, multicast)
• Adds TunnelDown to the Netlinker interface, called in each service's Teardown() before TunnelDelete

Why

When doublezerod deletes a tunnel, external applications (e.g. validators) may have sockets bound to the tunnel interface's overlay IP. If we delete the interface directly:

• TCP sockets: Existing connections get an abrupt RST. The peer knows the connection died, but the local application gets no advance warning.
• UDP sockets: The socket silently loses its binding. Sends may start using a different source address or fail with ENETUNREACH/EINVAL, but there's no explicit error delivered to the application — it just breaks quietly.

Setting the interface down first triggers ENETUNREACH errors on bound sockets immediately (the kernel processes LinkSetDown synchronously). Applications blocked on recv/epoll will see errors before the interface object is removed. This is a small improvement in behavior — not strictly necessary since the kernel does clean up everything on delete, but it turns silent breakage into explicit errors that are easier to debug.

The TunnelDown error is logged but not fatal — if the interface is already gone (e.g. double teardown), we don't want to block the TunnelDelete cleanup.

Testing Verification

• GOOS=linux go vet ./client/doublezerod/... passes
• All mock implementations updated to satisfy the extended Netlinker interface