builtin clangd integration is slower than using clangd as custom Language Client at highlighting syntax

Qt Creator can use custom LSP server in Language Client settings. I can set up clangd as the custom LSP server, with the custom arguments that matches mostly with the builtin one:

```
--compile-commands-dir=%

-j=8 --background-index --background-index-priority=low --header-insertion=never --limit-references=0 --rename-file-limit=0 --clang-tidy=0 --use-dirty-headers --ranking-model=decision_forest
```

And make sure the startup behavior is set to Start Server per Project.

Then open a project with precompiled headers enabled that contains many headers, like this:

```cmake
target_precompile_headers(
${TARGET_NAME}
PRIVATE

1. — stdlib —
   <string>
   <vector>
   <array>
   <map>
   <unordered_map>
   <set>
   <unordered_set>
   <queue>
   <stack>
   <deque>
   <ranges>
   <string_view>
   <tuple>
   <optional>
   <variant>
   <any>
   <bit>
   <iostream>
   <fstream>
   <sstream>
   <algorithm>
   <functional>
   <memory>
   <utility>
   <numeric>
   <limits>
   <mutex>
   <atomic>
   <thread>
   <chrono>
2. — Boost —
   <boost/asio.hpp>
   <boost/cobalt.hpp>
   <boost/container/stable_vector.hpp>
   <boost/container/static_vector.hpp>
   <boost/container/vector.hpp>
   <boost/hof.hpp>
   <boost/lambda2.hpp>
   <boost/move/utility.hpp>
   <boost/safe_numerics/exception.hpp>
   <boost/safe_numerics/safe_integer.hpp>
   <boost/safe_numerics/safe_integer_literal.hpp>
   <boost/safe_numerics/safe_integer_range.hpp>
   <boost/scope/defer.hpp>
   <boost/scope/scope_exit.hpp>
   <boost/scope/scope_fail.hpp>
   <boost/scope/scope_success.hpp>
   <boost/smart_ptr.hpp>
   <boost/smart_ptr/atomic_shared_ptr.hpp>
   <boost/smart_ptr/local_shared_ptr.hpp>
   <boost/smart_ptr/make_local_shared.hpp>
3. — Qt —
   <QGuiApplication>
   <QVulkanInstance>
   <QWindow>
   <QApplication>
   <QWidget>
   <QtCore>
   )
   ```

then the builtin clangd integration is much slower than the clangd running as custom LSP server in Language Client settings.

If I open a very small C++ source/header file, the custom clangd takes about 2 seconds to highlight the syntax, while the builtin one takes like 10 seconds.

I'm not sure about the reason but I suppose the PCH handling in Qt Creator is to be blamed. The only difference in arguments between those two is the `--compile-commands-dir`: the custom one points to the build folder where `compile_commands.json` is generated by CMake, while the builtin one points to the `.qtc_clang` directory.