feat: add model download screen and gate app on model presence

Adds a one-time download flow for the on-device Gemma 2B model (~1.1 GB)
with a progress bar, partial-download cleanup, and automatic navigation
to the app shell once the file is verified. App.dart now gates on
modelPresentProvider instead of onboarding state.

lib/core/llm/model_download_service.dart

@@ -0,0 +1,63 @@
+import 'dart:io';
+import 'package:path_provider/path_provider.dart';
+import 'package:path/path.dart' as p;
+
+class ModelDownloadService {
+  static const modelFileName = 'gemma-2b-it-cpu-int4.bin';
+  static const modelSizeBytes = 1183285248; // ~1.1 GB
+
+  // Gemma 2B IT CPU int4 — MediaPipe LLM Inference model
+  static const _downloadUrl =
+      'https://storage.googleapis.com/mediapipe-models/llm_inference/'
+      'gemma-2b-it-cpu-int4/float32/1/gemma-2b-it-cpu-int4.bin';
+
+  Future<String> get modelPath async {
+    final dir = await getApplicationSupportDirectory();
+    return p.join(dir.path, modelFileName);
+  }
+
+  Future<bool> isModelPresent() async {
+    final path = await modelPath;
+    final file = File(path);
+    if (!file.existsSync()) return false;
+    // Verify the file isn't a partial download (allow small variance)
+    final size = await file.length();
+    return size > modelSizeBytes * 0.99;
+  }
+
+  /// Streams download progress as a value from 0.0 to 1.0.
+  Stream<double> download() async* {
+    final path = await modelPath;
+    final file = File(path);
+    final client = HttpClient();
+
+    try {
+      final request = await client.getUrl(Uri.parse(_downloadUrl));
+      final response = await request.close();
+
+      if (response.statusCode != 200) {
+        throw Exception('Download failed: HTTP ${response.statusCode}');
+      }
+
+      final totalBytes =
+          response.contentLength > 0 ? response.contentLength : modelSizeBytes;
+      int receivedBytes = 0;
+
+      final sink = file.openWrite();
+      try {
+        await for (final chunk in response) {
+          sink.add(chunk);
+          receivedBytes += chunk.length;
+          yield (receivedBytes / totalBytes).clamp(0.0, 1.0);
+        }
+      } finally {
+        await sink.close();
+      }
+    } catch (_) {
+      if (await file.exists()) await file.delete();
+      rethrow;
+    } finally {
+      client.close();
+    }
+  }
+}