feat: add FOREMAN_KEEP_ALIVE config for worker model residency

Allow configuring how long the worker model stays resident on the Ollama
target after a request via FOREMAN_KEEP_ALIVE env var. Accepts Ollama
duration strings ("-1" forever, "0" unload, "15m", "1h", etc). Defaults
to "-1" (pin forever). The embedder warm-up is unaffected and always
uses keep_alive=-1.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

internal/worker/worker_test.go

@@ -40,7 +40,7 @@ func newTestWorker(t *testing.T, client ollama.Client) (*Worker, *store.Store, *
 	inv := ollama.NewModelInventory(client, logger)
 	notifier := NewNotifier()
 	dispatcher := webhook.NewDispatcher("", logger)
-	w := New(st, client, inv, notifier, dispatcher, logger)
+	w := New(st, client, inv, notifier, dispatcher, logger, "-1")
 	return w, st, notifier
 }
 
@@ -755,6 +755,74 @@ func TestStore_ResetInterruptedJobs(t *testing.T) {
 	}
 }
 
+func TestEncodeKeepAlive(t *testing.T) {
+	tests := []struct {
+		input string
+		want  string
+	}{
+		{"-1", "-1"},
+		{"0", "0"},
+		{"3600", "3600"},
+		{"15m", `"15m"`},
+		{"1h", `"1h"`},
+		{"", "-1"},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.input, func(t *testing.T) {
+			got := string(encodeKeepAlive(tt.input))
+			if got != tt.want {
+				t.Errorf("encodeKeepAlive(%q) = %s, want %s", tt.input, got, tt.want)
+			}
+		})
+	}
+}
+
+func TestWorker_SetsKeepAliveOnChatRequest(t *testing.T) {
+	client := &stubOllamaClient{}
+	st := openTestDB(t)
+	logger := slog.New(slog.NewJSONHandler(io.Discard, nil))
+	inv := ollama.NewModelInventory(client, logger)
+	notifier := NewNotifier()
+	dispatcher := webhook.NewDispatcher("", logger)
+
+	// Use "15m" to verify non-default keep_alive propagates to outbound requests.
+	w := New(st, client, inv, notifier, dispatcher, logger, "15m")
+
+	job := store.Job{
+		ID:          "01KEEPALIVE",
+		Model:       "qwen3:30b",
+		Payload:     json.RawMessage(`{"model":"qwen3:30b","messages":[{"role":"user","content":"hi"}]}`),
+		MaxAttempts: 3,
+	}
+	if _, err := st.CreateJob(job); err != nil {
+		t.Fatalf("CreateJob: %v", err)
+	}
+
+	waitCh := notifier.Register("01KEEPALIVE")
+
+	ctx, cancel := context.WithCancel(context.Background())
+	defer cancel()
+	go w.Run(ctx)
+
+	select {
+	case <-waitCh:
+	case <-time.After(5 * time.Second):
+		t.Fatal("timed out waiting for job to complete")
+	}
+
+	// Verify the chat request had keep_alive set to "15m".
+	client.mu.Lock()
+	defer client.mu.Unlock()
+	if len(client.chatCalls) != 1 {
+		t.Fatalf("expected 1 chat call, got %d", len(client.chatCalls))
+	}
+	gotKA := string(client.chatCalls[0].KeepAlive)
+	if gotKA != `"15m"` {
+		t.Errorf("keep_alive = %s, want %s", gotKA, `"15m"`)
+	}
+}
+
 func TestStore_DeleteTerminalJobsBefore(t *testing.T) {
 	st := openTestDB(t)