revert: move serve --bind fix out of K8s PR (to be submitted separately)

src/main.rs and entrypoint.sh bind address changes belong in a dedicated
fix PR. This PR should only contain K8s deployment configs and docs.

The deployment.yaml already handles the 127.0.0.1 limitation via the
hostPath /dev/kvm approach; users can add a socat sidecar if needed
until the fix PR is merged.

Author: Ubuntu

docker/entrypoint.sh

@@ -6,7 +6,6 @@ KERNEL="${ZEROBOOT_KERNEL:-${WORKDIR}/vmlinux-fc}"
 ROOTFS_PYTHON="${ZEROBOOT_ROOTFS_PYTHON:-${WORKDIR}/rootfs-python.ext4}"
 ROOTFS_NODE="${ZEROBOOT_ROOTFS_NODE:-}"
 PORT="${ZEROBOOT_PORT:-8080}"
-BIND="${ZEROBOOT_BIND:-0.0.0.0}"
 TEMPLATE_WAIT="${ZEROBOOT_TEMPLATE_WAIT:-15}"
 
 # ── Validate KVM access ───────────────────────────────────────────────────────
@@ -67,4 +66,4 @@ fi
 
 # ── Start API server ──────────────────────────────────────────────────────────
 echo "Starting zeroboot API server on port ${PORT}..."
-exec /usr/local/bin/zeroboot serve "$SERVE_TARGET" "$PORT" --bind "$BIND"
+exec /usr/local/bin/zeroboot serve "$SERVE_TARGET" "$PORT"

src/main.rs

@@ -30,7 +30,7 @@ fn main() -> Result<()> {
             eprintln!(
                 "  test-exec <workdir> <command>         - Test executing a command in a fork"
             );
-            eprintln!("  serve <workdir> [port] [--bind addr]  - Start API server (default bind: 0.0.0.0)");
+            eprintln!("  serve <workdir> [port]                - Start API server");
             Ok(())
         }
     }
@@ -426,26 +426,10 @@ fn load_api_keys() -> Vec<String> {
 
 fn cmd_serve(args: &[String]) -> Result<()> {
     if args.len() < 1 {
-        bail!("Usage: zeroboot serve <workdir>[,lang:workdir2,...] [port] [--bind <addr>]");
+        bail!("Usage: zeroboot serve <workdir>[,lang:workdir2,...] [port]");
     }
     let port: u16 = args.get(1).and_then(|p| p.parse().ok()).unwrap_or(8080);
 
-    // Parse optional --bind flag (default 0.0.0.0 for Kubernetes compatibility).
-    // K8s health probes and Service ClusterIP routing require the server to listen
-    // on all interfaces, not just localhost.
-    let bind_addr = {
-        let mut addr = "0.0.0.0".to_string();
-        let mut i = 2;
-        while i + 1 < args.len() {
-            if args[i] == "--bind" {
-                addr = args[i + 1].clone();
-                break;
-            }
-            i += 1;
-        }
-        addr
-    };
-
     // Parse workdir specs: "workdir" or "python:workdir1,node:workdir2"
     let mut templates = std::collections::HashMap::new();
     for spec in args[0].split(',') {
@@ -482,10 +466,10 @@ fn cmd_serve(args: &[String]) -> Result<()> {
             .route("/v1/metrics", axum::routing::get(metrics_handler))
             .with_state(state);
 
-        let listener = tokio::net::TcpListener::bind(format!("{}:{}", bind_addr, port))
+        let listener = tokio::net::TcpListener::bind(format!("127.0.0.1:{}", port))
             .await
             .unwrap();
-        eprintln!("Zeroboot API server listening on {}:{}", bind_addr, port);
+        eprintln!("Zeroboot API server listening on port {}", port);
         axum::serve(
             listener,
             app.into_make_service_with_connect_info::<std::net::SocketAddr>(),