Tentative fix for CUDA-10.2 windows build failures (pytorch#76204)

Summary:
`C10_UNUSED` somehow triggers segfault in some versions of NVCC that looks like something as follows:
```
caffe2\caffe2\operators\piecewise_linear_transform_op.h(65): internal error: assertion failed: gen_variable_decl: declared_type is NULL (cp_gen_be.c, line 22209 in gen_variable_decl)

1 catastrophic error detected in the compilation of "caffe2/caffe2/operators/piecewise_linear_transform_op.cu".
Compilation aborted.
nvcc error   : 'cudafe++' died with status 0xC0000409

```
Fixes regression introduced by  pytorch#75538 / D35747333 (pytorch@f6c275f)
Pull Request resolved: pytorch#76204

Test Plan: CI

Reviewed By: EscapeZero, atalman

Differential Revision: D35831451

fbshipit-source-id: f744d4688c9fd324f8f54b27781a3def97778d1e
(cherry picked from commit fd64655)

Author: malfet

caffe2/operators/deform_conv_op_impl.h

@@ -155,7 +155,8 @@ bool DeformConvOp<T, Context>::RunOnDeviceWithOrderNCHW() {
     col_buffer->Resize(buffer_shape);
     T* col_buffer_data = col_buffer->template mutable_data<T>();
     // Im2col, followed by gemm.
-    for (C10_UNUSED const auto image_id : c10::irange(N)) {
+    for (const auto image_id : c10::irange(N)) {
+      (void)image_id; // CUDA-10.2 on Windows crashes when C10_UNUSED macro is used
       for (const auto group_id : c10::irange(group_)) {
         DeformableIm2col(
             Xdata + group_id * input_offset,
@@ -342,7 +343,8 @@ bool DeformConvGradientOp<T, Context>::RunOnDeviceWithOrderNCHW() {
     math::Set<T, Context>(dX->numel(), 0, dXdata, &context_);
   }
 
-  for (C10_UNUSED const auto image_id : c10::irange(N)) {
+  for (const auto image_id : c10::irange(N)) {
+    (void)image_id; // CUDA-10.2 on Windows crashes when C10_UNUSED macro is used
     for (const auto group_id : c10::irange(group_)) {
       math::Gemm<T, Context>(
           CblasTrans,

caffe2/operators/piecewise_linear_transform_op.h

@@ -62,7 +62,8 @@ class PiecewiseLinearTransformOp final : public Operator<Context> {
       const int64_t num_bounds_per_group,
       const int64_t num_group) {
     const T* start = bounds;
-    for (C10_UNUSED const auto i : c10::irange(num_group)) {
+    for (const auto i : c10::irange(num_group)) {
+      (void)i; // CUDA-10.2 on Windows crashes when C10_UNUSED macro is used
       if (!std::is_sorted(start, start + num_bounds_per_group)) {
         return false;
       }