Support IPv4 link-local pod address allocation for egress NAT44 in IPv6-only pod networks

Is your proposed feature related to a problem?

The desire is to reduce the traffic through an external NAT64 device (and thus the transit cost) or, in some cases, remove the need for DNS64/NAT64.

This is a NAT44 alternative to #17878

Describe the feature you'd like

In clusters with IPv6-only pod networks, assign pods an IPv4 address out of 169.254.0.0/16. This address would allow outgoing IPv4 connections, performing NAT44 through one of two possible paths:

1. Through Egress Gateway. In this case the originating node could be IPv6-only.
2. Through bpf.masquerade=true. In this case, the originating node would have to be dual-stack.

It would be possible to support a hybrid mode, where egress to public IPv4 destination goes through Egress Gateway and egress to local IPv4 destinations go through bpf.masquerade=true. This complexity would not need to be in an initial implementation.

Notify relevant community channels

@cilium/sig-network

(Optional) Describe your proposed solution

No response