[QNN EP] Support quantized BatchNorm with per-channel DQ params on QNN HTP (#26959)

## Motivation:
QNN HTP was rejecting quantized BatchNorm models where parameters
(scale, mean, var) come through DequantizeLinear nodes with per-channel
INT8 quantization. This pattern is common in quantized models from
quantization tools.

## Changes:

- Helpers to resolve BatchNorm params through DQ nodes to their
underlying initializers
- Support per-channel dequantization for BatchNorm parameters
- Support input datatype of UFIXED_POINT_16
- Add unit test covering this QDQ params configuration

Author: qti-yuduo

onnxruntime/core/optimizer/qdq_transformer/selectors_actions/qdq_selectors.cc

@@ -758,7 +758,14 @@ bool BatchNormalizationNodeGroupSelector::Check(const GraphViewer& graph_viewer,
                                                 const Node* redundant_clip_node,
                                                 const std::vector<const Node*>& dq_nodes,
                                                 const std::vector<const Node*>& q_nodes) const {
-  if (!CheckQDQNodes(graph_viewer, node, redundant_clip_node, dq_nodes, q_nodes, 3)) {
+  // BatchNormalization has 5 inputs: x, scale, bias, mean, var.
+  // Require DQ on x and scale (indices 0,1). mean, var may optionally have DQ.
+  const int num_dq_nodes = gsl::narrow_cast<int>(dq_nodes.size());
+  if (num_dq_nodes < 3 || num_dq_nodes > 5) {
+    return false;
+  }
+
+  if (!CheckQDQNodes(graph_viewer, node, redundant_clip_node, dq_nodes, q_nodes, num_dq_nodes)) {
     return false;
   }