[net] close the listening socket on setup failure and open sockets CLOEXEC

[ManoharPaturi]

Problem

Two descriptor issues in getRecvSocket():

1. If fcntl(O_NONBLOCK) or setsockopt(SO_REUSEADDR) fails after the socket is created, the function returns -1 without closing the socket — the fd leaks for the lifetime of the supervisor. In --mode listen with --max_conns retry loops this accumulates.
2. The listening socket was created without SOCK_CLOEXEC, and acceptConn() accepted connections without SOCK_CLOEXEC as well, so both survive any execve() in the supervisor process (e.g. the newuidmap/newgidmap helper exec path in subproc::systemExe(), which has no fd sweep of its own).

Change

• close(sockfd) before the error returns after socket creation.
• SOCK_CLOEXEC on the socket() creation and on the accept4() flags.

Testing

Built and ran the full test suite (unit + cmdline) plus an ASan/UBSan build on Ubuntu 24.04 — all green, no new findings.

[ManoharPaturi]

Small branch hygiene fix: my local CI workflow file accidentally ended up in this branch — force-pushed the branch rebuilt from current master with only the intended change. The diff is now just the patch itself; the actions scan failure from earlier is gone.

[ManoharPaturi]

@robertswiecki Could you review this when convenient? It addresses two descriptor issues in getRecvSocket(): the listening socket was leaked when fcntl() or setsockopt() failed after creation, and both the listener and accepted connections were opened without SOCK_CLOEXEC. Happy to split the two changes into separate PRs if preferred.