[ET-VK][qconv2d][ez] Don't use im2col path for general convs

Pull Request resolved: pytorch#17393

This removes the dynamic dispatch logic in q8ta_conv2d() that selected between the im2col and general convolution paths. The function now unconditionally uses q8ta_conv2d_general(). This simplifies the dispatch since the im2col path selection will be handled upstream by the pattern matcher routing to specialized ops (q8ta_conv2d_pw, q8ta_conv2d_dw, etc.) instead of being decided at runtime.
ghstack-source-id: 341022577
@exported-using-ghexport

Differential Revision: [D93000164](https://our.internmc.facebook.com/intern/diff/D93000164/)

Author: ssjia

backends/vulkan/runtime/graph/ops/impl/Q8taConv2d.cpp

@@ -401,24 +401,7 @@ void q8ta_conv2d_general(
 }
 
 void q8ta_conv2d(ComputeGraph& graph, const std::vector<ValueRef>& args) {
-  // Index into args to extract values needed for dispatch decision
-  const ValueRef packed_int8_input = args.at(0);
-  const ValueRef kernel_size = args.at(9);
-  const ValueRef groups = args.at(13);
-
-  const int32_t groups_val = graph.get_int(groups);
-  const int64_t IC = graph.size_at<int64_t>(-3, packed_int8_input);
-
-  const int64_t K_h = graph.get_int_list(kernel_size)->at(0);
-  const int64_t K_w = graph.get_int_list(kernel_size)->at(1);
-
-  // Use im2col path when: non-grouped, input channels multiple of 4, small
-  // kernel
-  if (groups_val == 1 && IC % 4 == 0 && K_h <= 3 && K_w <= 3) {
-    q8ta_conv2d_im2col(graph, args);
-  } else {
-    q8ta_conv2d_general(graph, args);
-  }
+  q8ta_conv2d_general(graph, args);
 }
 
 REGISTER_OPERATORS {