Add patch for WebGPU on Android to handle fp16 in uniforms (microsoft#25349)

### Motivation and Context
Android devices (like S24) doesn't seem to allow fp16 in uniforms so the
WebGPU EP has to manually handle passing an fp32 in the uniform and
converting to fp16 before using.

Author: vraspar

cmake/external/onnxruntime_external_deps.cmake

@@ -747,7 +747,11 @@ if (onnxruntime_USE_WEBGPU)
           #
           # - (private) Fulfill the BinSkim requirements
           #   Some build warnings are not allowed to be disabled in project level.
-          ${Patch_EXECUTABLE} --binary --ignore-whitespace -p1 < ${PROJECT_SOURCE_DIR}/patches/dawn/dawn_binskim.patch)
+          ${Patch_EXECUTABLE} --binary --ignore-whitespace -p1 < ${PROJECT_SOURCE_DIR}/patches/dawn/dawn_binskim.patch &&
+
+          # Android devices doesn't seem to allow fp16 in uniforms so the WebGPU EP has to manually handle passing an fp32
+          # in the uniform and converting to fp16 before using.
+          ${Patch_EXECUTABLE} --binary --ignore-whitespace -p1 < ${PROJECT_SOURCE_DIR}/patches/dawn/uniform_and_storage_buffer_16_bit_access.patch)
 
       onnxruntime_fetchcontent_declare(
         dawn

cmake/patches/dawn/uniform_and_storage_buffer_16_bit_access.patch

@@ -0,0 +1,39 @@
+diff --git a/src/dawn/native/vulkan/DeviceVk.cpp b/src/dawn/native/vulkan/DeviceVk.cpp
+index c01d64e40f..0f1f4beae4 100644
+--- a/src/dawn/native/vulkan/DeviceVk.cpp
++++ b/src/dawn/native/vulkan/DeviceVk.cpp
+@@ -464,13 +464,15 @@ ResultOrError<VulkanDeviceKnobs> Device::CreateDevice(VkPhysicalDevice vkPhysica
+         DAWN_ASSERT(usedKnobs.HasExt(DeviceExt::ShaderFloat16Int8) &&
+                     mDeviceInfo.shaderFloat16Int8Features.shaderFloat16 == VK_TRUE &&
+                     usedKnobs.HasExt(DeviceExt::_16BitStorage) &&
+-                    mDeviceInfo._16BitStorageFeatures.storageBuffer16BitAccess == VK_TRUE &&
++                    mDeviceInfo._16BitStorageFeatures.storageBuffer16BitAccess == VK_TRUE /*&&
+                     mDeviceInfo._16BitStorageFeatures.uniformAndStorageBuffer16BitAccess ==
+-                        VK_TRUE);
++                        VK_TRUE*/);
+
+         usedKnobs.shaderFloat16Int8Features.shaderFloat16 = VK_TRUE;
+         usedKnobs._16BitStorageFeatures.storageBuffer16BitAccess = VK_TRUE;
+-        usedKnobs._16BitStorageFeatures.uniformAndStorageBuffer16BitAccess = VK_TRUE;
++        if (mDeviceInfo._16BitStorageFeatures.uniformAndStorageBuffer16BitAccess == VK_TRUE) {
++            usedKnobs._16BitStorageFeatures.uniformAndStorageBuffer16BitAccess = VK_TRUE;
++        }
+         if (mDeviceInfo._16BitStorageFeatures.storageInputOutput16 == VK_TRUE) {
+             usedKnobs._16BitStorageFeatures.storageInputOutput16 = VK_TRUE;
+         }
+diff --git a/src/dawn/native/vulkan/PhysicalDeviceVk.cpp b/src/dawn/native/vulkan/PhysicalDeviceVk.cpp
+index a324c101ed..8d64da750f 100644
+--- a/src/dawn/native/vulkan/PhysicalDeviceVk.cpp
++++ b/src/dawn/native/vulkan/PhysicalDeviceVk.cpp
+@@ -269,8 +269,9 @@ void PhysicalDevice::InitializeSupportedFeaturesImpl() {
+     if (mDeviceInfo.HasExt(DeviceExt::ShaderFloat16Int8) &&
+         mDeviceInfo.HasExt(DeviceExt::_16BitStorage) &&
+         mDeviceInfo.shaderFloat16Int8Features.shaderFloat16 == VK_TRUE &&
+-        mDeviceInfo._16BitStorageFeatures.storageBuffer16BitAccess == VK_TRUE &&
+-        mDeviceInfo._16BitStorageFeatures.uniformAndStorageBuffer16BitAccess == VK_TRUE) {
++        mDeviceInfo._16BitStorageFeatures.storageBuffer16BitAccess == VK_TRUE /*&&
++        WebGPU EP needs to ensure we don't put fp16 values in uniforms when this patch is applied.
++        mDeviceInfo._16BitStorageFeatures.uniformAndStorageBuffer16BitAccess == VK_TRUE*/) {
+         // ONNX Runtime Patch: enable shaderF16 on all devices.
+         EnableFeature(Feature::ShaderF16);
+         shaderF16Enabled = true;

tools/ci_build/github/android/default_full_aar_build_settings.json

@@ -16,6 +16,8 @@
         "--build_shared_lib",
         "--use_nnapi",
         "--use_xnnpack",
+        "--use_webgpu",
+        "--cmake_extra_defines=CMAKE_CXX_SCAN_FOR_MODULES=OFF",
         "--skip_tests"
     ]
 }