CI: fix unstable test \ext\sockets\tests\socket_recvfrom_afpacket_no_port.phpt

[LamentXU123]

Fix flaky CI: https://github.com/php/php-src/actions/runs/29266463698/job/86873055452

When investigating, I've noticed that similar tests does not have this issue. Only socket_recvfrom_afpacket_no_port.phpt is flaky. Which makes me discover a shared logic in tests:

https://github.com/php/php-src/blob/master/ext/sockets/tests/socket_sendto_recvfrom_afpacket_malformed.phpt#L28-L41

<?php
function recv_matching(Socket $s, string $header, int $maxlen = 65536, ?string &$addr = null, &$port = null): string|false {
    socket_set_nonblock($s);
    $deadline = microtime(true) + 5.0;
    while (microtime(true) < $deadline) {
        $bytes = @socket_recvfrom($s, $buf, $maxlen, 0, $addr, $port);
        if ($bytes !== false && is_string($buf) && str_starts_with($buf, $header)) {
            return $buf;
        }
        if ($bytes === false) {
            usleep(1000);
        }
    }
    return false;
}

This allows the test to wait for a frame matching its own Ethernet header before asserting the returned address, so it no longer depends on incidental packet ordering. ETH_P_ALL sockets on loopback can observe unrelated localhost traffic.

And only this single flaky test doesn't have that logic. Let's add them.

[devnexen]

nit: this file is the "no optional port argument" contract test, and the original $bytes >= 60 asserted the return value of the 5-arg call. The new version only checks strlen($buf), dropping that assertion. Equal by construction today, so nothing breaks, but since recv_matching() already has the count, consider surfacing it via a by-ref param to keep the $bytes >= 60 check.

[LamentXU123]

Should this be backported to 8.4?

[devnexen]

no. packet filter support did not exist yet.