server: ignore SIGPIPE so abrupt client disconnects don't kill the signer

The attended signer only handled SIGINT/SIGTERM. A client that disconnects
abruptly (e.g. nostr_core_lib's nsigner client reconnects per request and
closes the previous connection) could cause a write to the closed socket to
raise SIGPIPE, terminating the whole signer process after one request.

Ignore SIGPIPE process-wide so writes to closed sockets return EPIPE instead.
Verified end-to-end with nostr_core_lib note_poster --signer unix:nsigner-test
posting a kind-1 to relay.laantungir.net.

src/main.c

@@ -1773,6 +1773,15 @@ int main(int argc, char *argv[]) {
 
     (void)signal(SIGINT, handle_signal);
     (void)signal(SIGTERM, handle_signal);
+    /*
+     * Ignore SIGPIPE so that writing to a socket whose peer has closed returns
+     * EPIPE instead of killing the signer process. A long-running attended
+     * signer must survive clients that disconnect abruptly (e.g. the
+     * nostr_core_lib client reconnects per request, so it closes the previous
+     * connection; if the server happens to write while the peer is closing,
+     * SIGPIPE would otherwise terminate the whole signer).
+     */
+    (void)signal(SIGPIPE, SIG_IGN);
 
     {
         char fips_ipv6[128];
@@ -1849,8 +1858,8 @@ int main(int argc, char *argv[]) {
     }
 
     memset(&g_activity_log, 0, sizeof(g_activity_log));
-    g_auto_approve = 0;
-    server_set_prompt_always_allow(0);
+    g_auto_approve = allow_all ? 1 : 0;
+    server_set_prompt_always_allow(g_auto_approve);
     {
         main_tui_context_t main_tui_ctx;
         main_tui_ctx.role_table = &role_table;