fix(ports): handle startModelServer rejection on the promise, not a try/catch

SonarCloud reliability (C on new code): startModelServer became async (it scans for a
free port), so a try/catch wrapped around a fire-and-forget call can't catch its
rejection (index.ts:201). Handle it on the promise with .catch instead, matching the
llm.init().catch pattern beside it. Applied the same to the other two fire-and-forget
call sites (index.ts boot + ipc restart) so a rejected port scan is always logged, never
an unhandled promise rejection.

Author: alichherawalla

src/main/index.ts

@@ -198,11 +198,9 @@ app.whenReady().then(() => {
         /* ignore */
       }
     }
-    try {
-      void startModelServer()
-    } catch (e) {
-      console.error('[gateway] start failed', e)
-    }
+    // startModelServer is async (it scans for a free port); a try/catch can't catch its rejection,
+    // so handle it on the promise itself.
+    startModelServer().catch((e) => console.error('[gateway] start failed', e))
     void import('./llm').then(({ llm }) =>
       llm.init().catch((err) => console.error('[gateway] LLM init failed', err))
     )
@@ -329,7 +327,9 @@ app.whenReady().then(() => {
     setupIPC()
     setupRagIPC()
     setupMcpIpc() // basic MCP connectors (management + chat tool extension)
-    void startModelServer() // one OpenAI-compatible local gateway (LLM + STT); auto-picks a free port
+    // one OpenAI-compatible local gateway (LLM + STT); auto-picks a free port. Async, so handle a
+    // rejection on the promise (a try/catch around a fire-and-forget async call can't catch it).
+    startModelServer().catch((e) => console.error('[model-server] start failed', e))
     startMediaServer() // loopback HTTP for seekable local media (meeting videos)
     // Heal a stale active-model.json whose model gained a vision projector after it was
     // activated (e.g. Gemma 4 E2B) — turns vision on at launch if the projector is now

src/main/ipc.ts

@@ -1652,7 +1652,9 @@ export function setupIPC() {
       } catch {
         /* not running */
       }
-      void startModelServer() // re-listens; falls back to a free port if the preferred one is held
+      // re-listens; falls back to a free port if the preferred one is held. Async, so catch a
+      // rejection on the promise rather than leaving it unhandled.
+      startModelServer().catch((e) => console.error('[model-server] restart failed', e))
       return { success: true }
     }
     return { success: false, error: `cannot restart "${id}"` }
@@ -1984,9 +1986,8 @@ export function setupIPC() {
   // Which STT engine + model would run right now (provenance) + the installed transcription
   // models a picker can switch to (via the existing models:set-active-modal — this only lists).
   ipcMain.handle('transcription:active-info', async () => {
-    const { getActiveTranscriptionInfo, transcriptionModelOptions } = await import(
-      './transcription/select'
-    )
+    const { getActiveTranscriptionInfo, transcriptionModelOptions } =
+      await import('./transcription/select')
     const { listInstalled } = await import('./models-manager')
     const { modelsByKind } = await import('@offgrid/models')
     const info = getActiveTranscriptionInfo()