Optimize release builds: LTO, -O3, visibility, micro-perf fixes

- CMakeLists.txt: enable LTO for Release, add -O3 and -fvisibility=hidden
- scene: addMesh uses std::move, removeMesh takes const shared_ptr&
- entity: std::move entity into map instead of copy
- clouds: cosf/sinf instead of cos/sin (float math, avoids double promotion)
- game_screen: reserve trainer spell vector before push_back loop
- warden_module/emulator: replace std::endl (121 stream flushes) with '\n'

src/ui/game_screen.cpp

@@ -5112,6 +5112,7 @@ void GameScreen::renderTrainerWindow(game::GameHandler& gameHandler) {
             } else {
                 // Single tab or no categorization - flat list
                 std::vector<const game::TrainerSpell*> allSpells;
+                allSpells.reserve(trainer.spells.size());
                 for (const auto& spell : trainer.spells) {
                     allSpells.push_back(&spell);
                 }