fix: log vsock accept guard blocks reconnects after first client

[CodeMaster4711]

Root cause of logs staying empty in the frontend even after #213/#214 landed.

stream_logs in csfx-guest-init only calls listener.accept() while client.is_none(). Once one client connects, that branch of the select! never fires again until a write to the stale client fails — which only happens on the next log line produced. If the agent side reconnects (browser tab reload, gateway retry, etc.) before another log line is written, guest-init never pulls the new connection off the accept queue: the vsock handshake succeeds at the kernel level (agent logs guest log vsock connected), but nobody on the guest side ever reads or writes on it, so the stream sits open with zero bytes forever — exactly the symptom (log client accepted logged once, but the agent shows 5 separate connect attempts, only 1 of which was actually accepted).

Fix: drop the is_none() guard so a new connection always replaces whatever was there, dead or not.

[github-actions]

cargo audit found vulnerabilities:

• RUSTSEC-2026-0204 crossbeam-epoch@0.9.18: Invalid pointer dereference in fmt::Pointer impl for Atomic and Shared when the underlying pointer is invalid
• RUSTSEC-2026-0235 rkyv@0.7.46: Insufficient archive validation can cause out-of-bounds reads in archives containing Rc/Arc
• RUSTSEC-2023-0071 rsa@0.9.10: Marvin Attack: potential key recovery through timing sidechannels