feat: integrate Langfuse for LLM observability (#66)

* feat: integrate Langfuse for LLM observability

- Add instrumentation.ts with Langfuse OpenTelemetry exporter
- Enable experimental telemetry on streamText calls
- Add instrumentationHook to Next.js config
- Install required dependencies (@vercel/otel, langfuse-vercel, etc.)

* feat: add optional Langfuse observability integration

- Add session tracking with unique sessionId per conversation
- Add user tracking via IP address (x-forwarded-for header)
- Make telemetry conditional - only enabled if LANGFUSE_PUBLIC_KEY is set
- Add environment variable validation in instrumentation.ts
- Add sessionId validation (type check + 200 char limit)
- Update env.example with Langfuse configuration docs
- Remove unused langfuse-vercel and @vercel/otel packages

* fix: remove deprecated instrumentationHook (enabled by default in Next.js 15)

instrumentation.ts

@@ -0,0 +1,22 @@
+import { LangfuseSpanProcessor } from '@langfuse/otel';
+import { NodeTracerProvider } from '@opentelemetry/sdk-trace-node';
+
+export function register() {
+  // Skip telemetry if Langfuse env vars are not configured
+  if (!process.env.LANGFUSE_PUBLIC_KEY || !process.env.LANGFUSE_SECRET_KEY) {
+    console.warn('[Langfuse] Environment variables not configured - telemetry disabled');
+    return;
+  }
+
+  const langfuseSpanProcessor = new LangfuseSpanProcessor({
+    publicKey: process.env.LANGFUSE_PUBLIC_KEY,
+    secretKey: process.env.LANGFUSE_SECRET_KEY,
+    baseUrl: process.env.LANGFUSE_BASEURL,
+  });
+
+  const tracerProvider = new NodeTracerProvider({
+    spanProcessors: [langfuseSpanProcessor],
+  });
+
+  tracerProvider.register();
+}