Fix webgpu_native_test build after InputData input migration (#20439)

Summary:
`D108428753` migrated `WebGPUGraph::copy_inputs` and
`update_symints_from_inputs` to take `const std::vector<InputData>&`,
but the `select_as_symint` execute-read case in `test_webgpu_native.cpp`
still built its fake input list as `std::vector<std::pair<const void*,
size_t>>`, which has no conversion to `std::vector<InputData>`. As a
result `webgpu_native_test` failed to compile on `main`. This updates
that one call site to construct a `std::vector<InputData>` and pass
`host_is_int64=true` for the int64 `input_pos`, matching the landed
signature.

This diff was authored with assistance from Claude.

Differential Revision: D109396824

Author: JulianCloudNTH

backends/webgpu/test/test_webgpu_native.cpp

@@ -1490,12 +1490,11 @@ static bool test_symint_roundtrip(const std::string& blob_path) {
     return false;
   }
   const auto& in_ids = graph.input_ids();
-  std::vector<std::pair<const void*, size_t>> fake_inputs(
-      in_ids.size(), {nullptr, 0});
+  std::vector<InputData> fake_inputs(in_ids.size());
   int64_t fake_pos = 5;
   for (size_t i = 0; i < in_ids.size(); i++) {
     if (in_ids[i] == srcs[0].input_tensor_id) {
-      fake_inputs[i] = {&fake_pos, sizeof(int64_t)};
+      fake_inputs[i] = {&fake_pos, sizeof(int64_t), true};
     }
   }
   graph.update_symints_from_inputs(fake_inputs);