fix(liveness): skip self.close() on SIGTERM during HMR/watch dev mode

Deno's HMR and --watch modes use a parent/child process model where the
parent sends SIGTERM to signal the child to restart. The liveness
middleware's SIGTERM handler was calling self.close(), which kills the
process before Deno can complete its restart cycle — causing every dev
server to die on the first file change instead of hot-reloading.

Only call self.close() in production (k8s) where SIGTERM means the pod
must shut down. In dev mode (--unstable-hmr or --watch), let the
process terminate naturally so Deno can relaunch it.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Author: guitavano

runtime/middlewares/liveness.ts

@@ -65,12 +65,21 @@ const buildHandler = (
       }
     });
   };
+  // In production (k8s), SIGTERM comes from the kubelet and the process must
+  // shut down. In local dev, Deno's HMR/watch sends SIGTERM to the child
+  // process expecting a clean restart cycle — calling self.close() here kills
+  // the process before HMR can relaunch it, breaking hot-reload entirely.
+  const isHMR = Deno.args.some((a) => a.includes("--unstable-hmr"));
+  const isWatch = Deno.args.some((a) => a.includes("--watch"));
+  const isDev = isHMR || isWatch;
   try {
     if (Deno.build.os !== "windows") {
       Deno.addSignalListener("SIGTERM", () => {
         const checks = runChecks();
         console.log(checks);
-        self.close();
+        if (!isDev) {
+          self.close();
+        }
       });
     }
   } catch (err) {