[GPU] Gracefully handle zero batch size (#34515)

### Details:
- Avoid division by zero error when creating memory descriptor for GPU
plugin

### Tickets:
 - closes #24243 

### AI Assistance:
 - *AI assistance used: yes*
 - To generate tests

Author: mostafafaheem

src/plugins/intel_gpu/src/graph/impls/onednn/utils.cpp

@@ -351,12 +351,22 @@ class MemoryDescriptorBuilder {
         dnnl::memory::dims dims;
         auto fmt_tag = _target_fmt;
 
+        auto count_except = [this](size_t skip_idx) -> int64_t {
+            const auto& raw = _layout.get_tensor().raw;
+            int64_t result = 1;
+            for (size_t i = 0; i < raw.size(); ++i) {
+                if (i != skip_idx)
+                    result *= static_cast<int64_t>(raw[i]);
+            }
+            return result;
+        };
+
         if (fmt_tag == dnnl::memory::format_tag::ab && _flatten) {
             dims = flatten_tensor(_layout.get_tensor());
             dims.insert(dims.begin(), 1);
         } else if (fmt_tag == dnnl::memory::format_tag::ab) {
             dims.push_back(_layout.batch());
-            dims.push_back(_layout.get_tensor().count() / _layout.batch());
+            dims.push_back(count_except(0));
         } else if (fmt_tag == dnnl::memory::format_tag::abc) {
             dims.push_back(_layout.batch());
             dims.push_back(_layout.feature());
@@ -385,7 +395,7 @@ class MemoryDescriptorBuilder {
             dims.push_back(_layout.spatial(1));
         } else if (fmt_tag == dnnl::memory::format_tag::ba) {
             dims.push_back(_layout.feature());
-            dims.push_back(_layout.get_tensor().count() / _layout.feature());
+            dims.push_back(count_except(1));
         } else if (_flatten) {
             dims = flatten_tensor(_layout.get_tensor());
         } else {

src/plugins/intel_gpu/tests/unit/onednn/utils_test.cpp

@@ -533,3 +533,29 @@ TEST_F(test_layout_to_memory_desc, regression_3d_shape_format_selection) {
     EXPECT_EQ(get_format_tag_from_desc(desc_byxf), dnnl::memory::format_tag::acb);
     // Format tags should be different (abc vs acb)
 }
+
+TEST_F(test_layout_to_memory_desc, zero_batch_ab_format) {
+    // batch=0 should not cause division-by-zero; inner dims should be preserved
+    layout l = layout{ov::PartialShape{0, 256}, data_types::f32, format::bfyx};
+    auto desc = layout_to_memory_desc(l, dnnl::memory::format_tag::ab);
+    EXPECT_EQ(desc.get_ndims(), 2);
+    EXPECT_EQ(desc.get_dims()[0], 0);
+    EXPECT_EQ(desc.get_dims()[1], 256);
+}
+
+TEST_F(test_layout_to_memory_desc, zero_feature_ba_format) {
+    // feature=0 should not cause division-by-zero; inner dims should be preserved
+    layout l = layout{ov::PartialShape{256, 0}, data_types::f32, format::bfyx};
+    auto desc = layout_to_memory_desc(l, dnnl::memory::format_tag::ba);
+    EXPECT_EQ(desc.get_ndims(), 2);
+    EXPECT_EQ(desc.get_dims()[0], 0);
+    EXPECT_EQ(desc.get_dims()[1], 256);
+}
+
+TEST_F(test_layout_to_memory_desc, zero_batch_default_format) {
+    // All-zero batch with default format should not crash
+    layout l = layout{ov::PartialShape{0, 64, 32, 32}, data_types::f16, format::bfyx};
+    EXPECT_NO_THROW({
+        auto desc = layout_to_memory_desc(l, dnnl::memory::format_tag::undef);
+    });
+}