[GPU] Code refactoring for cloning fc primitive (#30391)

### Details:
- Previously, we had member-by-member copy for bias fusion, which is
error-prone
 - Added clone() method to copy FC primitive
- Removed const from primitive id because it can be changed after
creation

---------

Co-authored-by: Sergey Shlyapnikov <[email protected]>

Author: isanghao

src/plugins/intel_gpu/include/intel_gpu/primitives/primitive.hpp

@@ -208,7 +208,7 @@ struct primitive {
     const primitive_type_id type;
 
     /// @brief Primitive's id.
-    const primitive_id id;
+    primitive_id id;
 
     /// @brief Name of original ov operation.
     std::string origin_op_name;
@@ -259,7 +259,7 @@ struct primitive {
         std::string type_str;
         ib >> type_str;
         *const_cast<primitive_type_id*>(&type) = prim_map_storage::instance().get_type_id(type_str);
-        ib >> *const_cast<primitive_id*>(&id);
+        ib >> id;
         ib >> origin_op_name;
         ib >> origin_op_type_name;
         ib >> output_paddings;
@@ -306,6 +306,9 @@ struct primitive {
 /// @brief base class for all primitives implementations.
 template <class PType>
 class primitive_base : public primitive {
+public:
+    std::shared_ptr<PType> clone() const { return std::make_shared<PType>(static_cast<const PType &>(*this)); }
+
 protected:
     explicit primitive_base(const primitive_id& id,
                             const std::vector<input_info>& input,

src/plugins/intel_gpu/src/graph/graph_optimizer/prepare_primitive_fusing.cpp

@@ -462,25 +462,11 @@ void prepare_primitive_fusing::fuse_bias(program &p) {
                 continue;
             }
 
-            auto fc_with_bias_prim = std::make_shared<fully_connected>(desc->id + "_tmp",
-                                                                       desc->input[0],
-                                                                       desc->weights,
-                                                                       bias_name,
-                                                                       fc.get_output_layout().data_type,
-                                                                       desc->input_size,
-                                                                       desc->weights_rank);
-
-            if (desc->compressed_weights) {
-                fc_with_bias_prim->compressed_weights = true;
-                fc_with_bias_prim->decompression_scale = desc->decompression_scale;
-                fc_with_bias_prim->decompression_zero_point = desc->decompression_zero_point;
-                if (desc->decompression_zero_point_scalar.has_value())
-                    fc_with_bias_prim->decompression_zero_point_scalar = desc->decompression_zero_point_scalar.value();
-                fc_with_bias_prim->activation_scale = desc->activation_scale;
-                fc_with_bias_prim->activation_zero_point = desc->activation_zero_point;
-                fc_with_bias_prim->dynamic_quantized_activation = desc->dynamic_quantized_activation;
-                fc_with_bias_prim->dynamic_quantized_activation_zp = desc->dynamic_quantized_activation_zp;
-            }
+            auto fc_with_bias_prim = desc->clone();
+            fc_with_bias_prim->id = desc->id + "_tmp";
+            fc_with_bias_prim->bias = bias_name;
+            fc_with_bias_prim->output_data_types = {optional_data_type{fc.get_output_layout().data_type}};
+
             auto& new_fc_node = p.get_or_create(fc_with_bias_prim);
             fuse_bias_f(fc, new_fc_node, bias_node, eltw_node);
         }

src/plugins/intel_gpu/src/graph/program.cpp

@@ -949,16 +949,12 @@ void program::rename(program_node& node, primitive_id const& new_id) {
     nodes_map.emplace(new_id, node_ptr);
     nodes_map.erase(node.id());
 
-    const_cast<primitive_id&>(node.desc->id) = new_id;
+    node.desc->id = new_id;
 }
 
 void program::swap_names(program_node& node1, program_node& node2) {
-    const auto _extract_id = [](program_node& node) -> primitive_id& {
-        return const_cast<primitive_id&>(node.desc->id);
-    };
-
     nodes_map.at(node1.id()).swap(nodes_map.at(node2.id()));
-    std::swap(_extract_id(node1), _extract_id(node2));
+    std::swap(node1.desc->id, node2.desc->id);
 }
 
 void program::replace_all_usages(program_node& old_node, program_node& new_node, bool remove_if_dangling) {
@@ -1026,8 +1022,8 @@ void program::replace(program_node& old_node, program_node& new_node) {
     new_node.constant = old_node.constant;
     new_node.data_flow = old_node.data_flow;
     new_node.user_mark = old_node.user_mark;
-    const_cast<std::string&>(new_node.desc->origin_op_name) = old_node.desc->origin_op_name;
-    const_cast<std::string&>(new_node.desc->origin_op_type_name) = old_node.desc->origin_op_type_name;
+    new_node.desc->origin_op_name = old_node.desc->origin_op_name;
+    new_node.desc->origin_op_type_name = old_node.desc->origin_op_type_name;
 
     processing_order.insert(&old_node, &new_node);
     if (processing_order.get_processing_iterator(old_node) != processing_order.end())