Inconsistent binding on Windows host

[pdematteis]

Describe the bug

In a Zenoh application, if no listen endpoint is specified, both IPv4 and IPv6 addresses should be used and available for peers to connect to.
Running the z_scout demo, it would report something like:
Hello { zid: Some(B063BC6F09C357450E94F8AEA620E83B), whatami: "Peer", locators: ["tcp/[fe80::55e6:c657:fa8a:d9ee]:64033", "tcp/192.168.1.51:64033"] }

Everything looks good right now, but when trying to connect, only IPv6 locators work.
Checking the TCP/UDP ports status with TCPView does confirm that only IPv6 addresses are used.

This happens only on Windows hosts, while Linux hosts work as expected.
It is true for both my own applications and for the precompiled zenohd router.

If the same locators are explicitly added to the session JSON configuration, like for example with:
{mode: "peer", listen: {endpoints: ["tcp/[fe80::55e6:c657:fa8a:d9ee]:64033", "tcp/192.168.1.51:64033"]}}
both IPv4 and IPv6 addresses work

Using JSON and specifying any address for both IPv4 and IPv6 also work:
{mode: "peer", listen: {endpoints: ["tcp/0.0.0.0:0", "tcp/[::]:0"]}}

To reproduce

1. Start zenohd on Windows and take note of the available locators
2. Use a client mode publisher and try to connect to all of them one by one
3. Do the same on Linux

System info

• Windows 11
• Debian 11