[mdns]: Minor corrections to mdns socket layer

[david-cermak]

mDNS socket layer cleanup (mdns_networking_socket.c)

This note documents follow-up fixes applied to the BSD-socket based mDNS networking backend (components/mdns/mdns_networking_socket.c).

1. Deterministic RX task shutdown (remove fixed delay)

Problem: mdns_priv_if_deinit() previously stopped the RX task with a fixed vTaskDelay(500ms), which is not a deterministic synchronization mechanism (task may still be running; rapid re-init can race).

Correction:

• Track the RX task using s_sock_recv_task_handle.
• Add a binary semaphore s_sock_recv_task_exit_sem.
• On last-interface deinit:
  • set s_run_sock_recv_task = false
  • block on s_sock_recv_task_exit_sem (bounded wait)
  • clear the task handle
• In the RX task: xSemaphoreGive(s_sock_recv_task_exit_sem) immediately before vTaskDelete(NULL).

2. Improve select()/recvfrom() unblocking during socket close

Problem: Closing a socket from another task does not reliably and immediately unblock a thread blocked in select()/recvfrom() on all stacks/ports.

Correction:

• delete_socket() now performs a best-effort shutdown(sock, SHUT_RDWR) prior to close(sock) to encourage prompt unblocking.

3. recvfrom() error handling (avoid repeated error/log-spin)

Problem: A recvfrom() error only broke out of the inner per-interface loop, leaving the RX task alive and potentially re-hitting the same invalid socket each select cycle.

Correction:

• On recvfrom() failure:
  • log at WARN
  • close the socket (delete_socket(sock))
  • set s_interfaces[tcpip_if].sock = -1
  • clear s_interfaces[tcpip_if].proto = 0
  • continue with other interfaces
• select() now ignores EINTR by retrying the loop.

4. Multicast group join failure must not mark protocol as ready

Problem: create_pcb() set the per-interface protocol bit even when join_mdns_multicast_group() failed, resulting in an interface being treated as ready for that protocol while multicast RX might be non-functional.

Correction:

• If join_mdns_multicast_group() fails, create_pcb() returns false and does not set the protocol bit.
• If the socket was newly created for this interface and no other protocol is active, it is closed to avoid leaking a socket with no valid membership.

5. Normalize socket close paths

Problem: Socket cleanup was split between direct close() calls and delete_socket(), risking divergence if delete_socket() gains additional behavior.

Correction:

• create_socket() error path now calls delete_socket(sock) instead of close(sock).