Fix array-bounds error (pytorch#3798)

Summary:
X-link: facebookresearch/FBGEMM#1485

Fix the following g++ error:
```
  In member function ‘constexpr const std::array<_Tp, _Nm>::value_type& std::array<_Tp, _Nm>::operator[](size_type) const [with _Tp = int; long unsigned int _Nm = 1]’,
      inlined from ‘fbgemm::GenConvKernelBase<SPATIAL_DIM, INST_SET>::GenConvKernelBase(const fbgemm::conv_param_t<kSpatialDim>&, int32_t, bool, bool, bool, bool, bool) [with int SPATIAL_DIM = 1; fbgemm::inst_set_t INST_SET = fbgemm::inst_set_t::avx512]’ at ../third_party/fbgemm/src/./GroupwiseConv.h:128:43:
  /usr/include/c++/14.2.1/array:219:24: error: array subscript 18446744073709551615 is above array bounds of ‘std::__array_traits<int, 1>::_Type’ {aka ‘const int [1]’} [-Werror=array-bounds=]
    219 |         return _M_elems[__n];
        |                ~~~~~~~~^
  /usr/include/c++/14.2.1/array: In constructor ‘fbgemm::GenConvKernelBase<SPATIAL_DIM, INST_SET>::GenConvKernelBase(const fbgemm::conv_param_t<kSpatialDim>&, int32_t, bool, bool, bool, bool, bool) [with int SPATIAL_DIM = 1; fbgemm::inst_set_t INST_SET = fbgemm::inst_set_t::avx512]’:
  /usr/include/c++/14.2.1/array:115:55: note: while referencing ‘std::array<int, 1>::_M_elems’
    115 |       typename __array_traits<_Tp, _Nm>::_Type        _M_elems;
        |                                                       ^~~~~~~~
  In member function ‘constexpr const std::array<_Tp, _Nm>::value_type& std::array<_Tp, _Nm>::operator[](size_type) const [with _Tp = int; long unsigned int _Nm = 1]’,
      inlined from ‘fbgemm::GenConvKernelBase<SPATIAL_DIM, INST_SET>::GenConvKernelBase(const fbgemm::conv_param_t<kSpatialDim>&, int32_t, bool, bool, bool, bool, bool) [with int SPATIAL_DIM = 1; fbgemm::inst_set_t INST_SET = fbgemm::inst_set_t::avx512_vnni]’ at ../third_party/fbgemm/src/./GroupwiseConv.h:128:43:
  /usr/include/c++/14.2.1/array:219:24: error: array subscript 18446744073709551615 is above array bounds of ‘std::__array_traits<int, 1>::_Type’ {aka ‘const int [1]’} [-Werror=array-bounds=]
    219 |         return _M_elems[__n];
        |                ~~~~~~~~^
  /usr/include/c++/14.2.1/array: In constructor ‘fbgemm::GenConvKernelBase<SPATIAL_DIM, INST_SET>::GenConvKernelBase(const fbgemm::conv_param_t<kSpatialDim>&, int32_t, bool, bool, bool, bool, bool) [with int SPATIAL_DIM = 1; fbgemm::inst_set_t INST_SET = fbgemm::inst_set_t::avx512_vnni]’:
  /usr/include/c++/14.2.1/array:115:55: note: while referencing ‘std::array<int, 1>::_M_elems’
    115 |       typename __array_traits<_Tp, _Nm>::_Type        _M_elems;
        |                                                       ^~~~~~~~
  cc1plus: all warnings being treated as errors
```
There was an out of bound access before fix, and that access was shortcuted by ``and`` condition. Nevertheless, it's better to fix that.

Pull Request resolved: pytorch#3798

Reviewed By: gchalump

Differential Revision: D77493350

Pulled By: q10

fbshipit-source-id: 285200a3bc69ff42fb53ff4f05a60e3961927fdb

Author: cyyever

src/GroupwiseConv.h

@@ -124,8 +124,12 @@ class GenConvKernelBase {
     H_PAD_ = conv_param.pad[0];
     W_PAD_ = conv_param.pad[1];
 
-    use_bottom_padding_ =
-        !(STRIDE_ > 1 && conv_param.IN_DIM[SPATIAL_DIM - 2] % 2 == 0);
+    if constexpr (SPATIAL_DIM >= 2) {
+      use_bottom_padding_ =
+          !(STRIDE_ > 1 && conv_param.IN_DIM[SPATIAL_DIM - 2] % 2 == 0);
+    } else {
+      use_bottom_padding_ = false;
+    }
     use_right_padding_ =
         !(STRIDE_ > 1 && conv_param.IN_DIM[SPATIAL_DIM - 1] % 2 == 0);
   }