Update oneDNN submodule and remove cache cleanup WA

Author: sshlyapn

src/plugins/intel_gpu/include/intel_gpu/plugin/plugin.hpp

@@ -45,7 +45,6 @@ class Plugin : public ov::IPlugin {
 
 public:
     Plugin();
-    ~Plugin();
 
     std::shared_ptr<ov::ICompiledModel> compile_model(const std::shared_ptr<const ov::Model>& model,
                                                       const ov::AnyMap& properties) const override;

src/plugins/intel_gpu/src/plugin/plugin.cpp

@@ -158,21 +158,6 @@ Plugin::Plugin() {
     m_compiled_model_runtime_properties["OV_VERSION"] = ov_version.buildNumber;
 }
 
-Plugin::~Plugin() {
-#ifdef ENABLE_ONEDNN_FOR_GPU
-    // To prevent hanging during oneDNN's primitive cache desctruction,
-    // trigger earlier cache cleanup by setting its capacity to 0.
-    // Related ticket: 106154.
-    dnnl::set_primitive_cache_capacity(0);
-
-    // In case of multiple ov::Core instances (and multiple GPU plugins) we need to restore original
-    // cache capacity to prevent working with zero-capacity cache in other GPU Plugin instances, since
-    // cache is shared between all of GPU Plugin instances and cache clean up affects all of them.
-    const int default_cache_capacity = 1024;
-    dnnl::set_primitive_cache_capacity(default_cache_capacity);
-#endif
-}
-
 std::shared_ptr<ov::ICompiledModel> Plugin::compile_model(const std::shared_ptr<const ov::Model>& model, const ov::AnyMap& orig_config) const {
     OV_ITT_SCOPED_TASK(itt::domains::intel_gpu_plugin, "Plugin::compile_model");
     std::string device_id = get_device_id(orig_config);