[GPU] Invalidate cache if it contains patchtoken binaries and driver version changed (#29195)

### Tickets:
 - https://jira.devtools.intel.com/browse/CVS-159917

---------

Co-authored-by: Pavel Durandin <pavel.durandin@intel.com>

Author: tkrupa-intel

src/plugins/intel_gpu/src/graph/impls/ocl/kernels_cache.cpp

@@ -636,25 +636,56 @@ void kernels_cache::add_to_cached_kernels(const std::vector<kernel::ptr>& kernel
 
 void kernels_cache::save(BinaryOutputBuffer& ob) const {
     ob << _cached_binaries.size();
+
+    auto is_zebin = [](const std::vector<unsigned char>& bin) {
+        constexpr uint32_t ELF_MAGIC = 0x464C457F;
+
+        if (bin.size() < sizeof(uint32_t)) {
+            return false;
+        }
+        auto magic = reinterpret_cast<const uint32_t*>(bin.data())[0];
+        return magic == ELF_MAGIC;
+    };
+
     for (auto& cached_binary : _cached_binaries) {
+        auto is_zebin_binary = is_zebin(cached_binary.first);
+
         ob << cached_binary.second;
         ob << cached_binary.first;
+        ob << is_zebin_binary;
+        if (!is_zebin_binary) {
+            auto driver_version = downcast<ocl::ocl_device>(*_device).get_info().driver_version;
+            ob << driver_version;
+        }
     }
 }
 
 void kernels_cache::load(BinaryInputBuffer& ib) {
     std::unordered_map<uint32_t, std::vector<unsigned char>> precompiled_kernels;
 
+    const auto& build_device = downcast<ocl::ocl_device>(*_device);
+
     size_t num_cached_binaries;
     ib >> num_cached_binaries;
     for (size_t i = 0; i < num_cached_binaries; ++i) {
+        bool is_zebin_binary = true;
+
         uint32_t id;
         ib >> id;
         ib >> precompiled_kernels[id];
+        ib >> is_zebin_binary;
+        if (!is_zebin_binary) {
+            // Legacy patchtoken path
+            std::string driver_version, current_driver_version;
+            ib >> driver_version;
+            current_driver_version = build_device.get_info().driver_version;
+
+            if (driver_version != current_driver_version) {
+                OPENVINO_THROW("Driver version mismatch in cached patchtoken kernels");
+            }
+        }
     }
 
-    const auto& build_device = downcast<ocl::ocl_device>(*_device);
-
     try {
         std::lock_guard<std::mutex> lock(_mutex);
         _cached_kernels.clear();

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

@@ -375,13 +375,9 @@ ov::Any Plugin::get_property(const std::string& name, const ov::AnyMap& options)
 
     ov::AnyMap actual_runtime_info;
     auto prepare_actual_runtime_info = [&]() {
-        // Suppose all devices share the same version driver.
-        auto device_id = m_default_device_id;
-        OPENVINO_ASSERT(m_device_map.find(device_id) != m_device_map.end(),
-                        "[GPU] compiled_model_runtime_properties: Couldn't find device for GPU with id ",
-                        device_id);
-        actual_runtime_info["DRIVER_VERSION"] = m_device_map.at(device_id)->get_info().driver_version;
-        // More items can be inserted if needed
+        // Items can be inserted here if needed.
+        // Note: driver version used to be here, but it was moved to cache load logic. See
+        // https://github.com/openvinotoolkit/openvino/pull/29195
     };
     // Below properties depend on the device ID.
     if (name == ov::internal::compiled_model_runtime_properties.name()) {