feat: wire real LLM runner via third-one or OpenAI-compatible API

internal/app/server.go

@@ -11,6 +11,7 @@ import (
 	"tutor/internal/httpapi"
 	"tutor/internal/interview"
 	"tutor/internal/learnermemory"
+	"tutor/internal/llm"
 	"tutor/internal/ontology"
 	"tutor/internal/progression"
 	"tutor/internal/teachingassets"
@@ -18,7 +19,15 @@ import (
 )
 
 func NewServer(cfg config.Config) *http.Server {
-	runner := workflows.NewStubRunner()
+	var runner workflows.Runner
+	if cfg.HasLLM() {
+		client := llm.NewClient(cfg.LLMEndpoint, cfg.LLMAPIKey, cfg.ModelKey)
+		runner = workflows.NewLLMRunner(client)
+		log.Printf("using llm runner: endpoint=%s model=%s", cfg.LLMEndpoint, cfg.ModelKey)
+	} else {
+		runner = workflows.NewStubRunner()
+		log.Println("using stub runner (TUTOR_LLM_ENDPOINT not set)")
+	}
 
 	var interviewStore interview.Store
 	var memoryStore learnermemory.Store