Use std::unique_ptr for DeviceMesh and FusionExecutorCache in direct bindings

[rdspring1]

Returning std::unique_ptr explicitly signals exclusive ownership transfer to Python.
Returning new use the default policy for pointer, which is return_value_policy::take_ownership.

[rdspring1]

!test

[github-actions]

Description

• Use std::make_unique for safe ownership transfer

• Replace raw new with smart pointers in bindings

• Update version to reflect changes

---

Changes walkthrough 📝

	Relevant files
Enhancement	multidevice.cpp Use unique_ptr for DeviceMesh in bindings                                python/python_direct/multidevice.cpp Replace new DeviceMesh with std::make_unique Maintain backward compatibility for integer list input +2/-2      runtime.cpp Use unique_ptr for FusionExecutorCache                                      python/python_direct/runtime.cpp Replace new FusionExecutorCache with std::make_unique Ensure exclusive ownership transfer to Python +1/-1
Configuration changes	version.txt Update version number                                                                        python/version.txt Bump version from 0.2.34 to 0.2.35 +1/-1

---

PR Reviewer Guide 🔍

Here are some key observations to aid the review process:

🧪 No relevant tests

⚡ Recommended focus areas for review Ownership Clarity The use of std::make_unique correctly transfers exclusive ownership to Python, but it should be confirmed that the pybind11 bindings properly handle std::unique_ptr return types to avoid memory leaks or undefined behavior. return std::make_unique<DeviceMesh>(std::move(devices)); Unique Pointer Initialization The nested use of std::make_unique for both FusionExecutorCache and Fusion should be validated to ensure correct object construction and ownership transfer, particularly under error or exception conditions. return std::make_unique<FusionExecutorCache>( std::make_unique<Fusion>(*fusion), fusion_id, auto_schedule);