feat: add gallery service and ML depth estimation placeholder

Media library access via new expo-media-library Query API,
image resize for wallpaper export, and MiDaS depth map
service stub for V1.5.

src/services/media/gallery.service.ts

@@ -0,0 +1,50 @@
+import * as MediaLibrary from "expo-media-library";
+import * as ImageManipulator from "expo-image-manipulator";
+import { Dimensions } from "react-native";
+
+const { width: SCREEN_WIDTH, height: SCREEN_HEIGHT } = Dimensions.get("screen");
+
+export async function requestGalleryPermission(): Promise<boolean> {
+  const { status } = await MediaLibrary.requestPermissionsAsync();
+  return status === "granted";
+}
+
+export async function getRecentPhotos(
+  count: number = 50,
+  offset: number = 0
+): Promise<{ assets: MediaLibrary.Asset[]; hasMore: boolean }> {
+  const query = new MediaLibrary.Query()
+    .eq(MediaLibrary.AssetField.MEDIA_TYPE, MediaLibrary.MediaType.IMAGE)
+    .orderBy({
+      key: MediaLibrary.AssetField.CREATION_TIME,
+      ascending: false,
+    })
+    .limit(count)
+    .offset(offset);
+
+  const assets = await query.exe();
+
+  return {
+    assets,
+    hasMore: assets.length === count,
+  };
+}
+
+export async function prepareImageForWallpaper(
+  uri: string
+): Promise<string> {
+  const result = await ImageManipulator.manipulateAsync(
+    uri,
+    [
+      {
+        resize: {
+          width: SCREEN_WIDTH * 2,
+          height: SCREEN_HEIGHT * 2,
+        },
+      },
+    ],
+    { compress: 0.95, format: ImageManipulator.SaveFormat.JPEG }
+  );
+
+  return result.uri;
+}

src/services/ml/depth.service.ts

@@ -0,0 +1,46 @@
+import { Platform } from "react-native";
+
+// Depth estimation service — V1.5 feature
+// Uses MiDaS v2.1 small (TFLite) for on-device depth map generation
+// Model size: ~17MB, inference: ~50-80ms on mid-range devices
+
+export interface DepthMapResult {
+  uri: string; // Path to grayscale depth map PNG
+  width: number;
+  height: number;
+  inferenceTimeMs: number;
+}
+
+export async function isDepthEstimationAvailable(): Promise<boolean> {
+  // Will check for TFLite model availability
+  // For now, return false — not yet implemented
+  return false;
+}
+
+export async function generateDepthMap(
+  imageUri: string
+): Promise<DepthMapResult> {
+  // TODO V1.5: Implement with react-native-fast-tflite or ONNX Runtime
+  //
+  // Pipeline:
+  // 1. Load image, resize to MiDaS input size (256x256 or 384x384)
+  // 2. Normalize to [0,1] float tensor
+  // 3. Run MiDaS inference
+  // 4. Resize output to original image dimensions
+  // 5. Normalize depth values to [0,255] grayscale
+  // 6. Save as PNG, return URI
+  //
+  // The depth map will be cached in MMKV keyed by image URI hash
+  // to avoid recomputation.
+
+  throw new Error(
+    "Depth estimation not yet available. Coming in V1.5 with MiDaS on-device inference."
+  );
+}
+
+export async function downloadDepthModel(): Promise<void> {
+  // TODO: Download MiDaS TFLite model from Cloudflare R2
+  // Model is not bundled in the app to keep binary size down
+  // Download on first use of depth-parallax animation
+  throw new Error("Model download not yet implemented.");
+}