make quantized_max_pool2d_nhwc handle case of C>64 (#19238)

Summary:

now the TIE quantized_max_pool2d_nhwc general path processes channels in chunks of 16 groups (64 bytes) at a time using a fixed stack array with an outer loop. This supports arbitrary C (any multiple of 4).
Also adds test cases for C=128, C=256, k=3x3, and padding to cover all TIE kernel dispatch paths.

Reviewed By: khazaei

Differential Revision: D103096179

Author: wl1026sun

backends/cadence/hifi/operators/operators.h

@@ -226,6 +226,16 @@ ::executorch::aten::Tensor& transpose_copy_int_out(
     int64_t dim1,
     ::executorch::aten::Tensor& out);
 
+::executorch::aten::Tensor& quantized_max_pool2d_nhwc_out(
+    ::executorch::runtime::KernelRuntimeContext& ctx,
+    const ::executorch::aten::Tensor& input,
+    ::executorch::aten::IntArrayRef kernel_size,
+    ::executorch::aten::IntArrayRef stride,
+    ::executorch::aten::IntArrayRef padding,
+    ::executorch::aten::IntArrayRef dilation,
+    bool ceil_mode,
+    ::executorch::aten::Tensor& output);
+
 } // namespace native
 } // namespace HiFi
 } // namespace impl