Implement pytest for 03_all_store/all_store_bench.py with kernel-level testing

[Copilot]

This PR adds a pytest test file for the all-store benchmark that focuses on testing the core kernel functionality rather than the full benchmarking infrastructure.

Implementation

The test follows the same pattern as test_load_bench.py by:

• Testing the kernel directly: Calls all_store_kernel instead of the full run_experiment() function
• Avoiding synchronization issues: Removes barrier calls that were causing Fatal Python errors in CI
• Parametrized testing: Covers multiple data types (int8, float16, bfloat16, float32), buffer sizes, and block sizes
• Minimal resource usage: Tests core functionality without expensive benchmarking overhead

Why Kernel-Level Testing

The original approach of calling run_experiment() caused crashes due to GPU synchronization barriers in the CI environment. By testing the all_store_kernel directly, we:

• Validate the core RMA store functionality
• Avoid environment-specific synchronization issues
• Follow the established pattern from test_load_bench.py
• Maintain fast, reliable test execution

The test ensures the all-store kernel can be invoked with various data types and configurations, providing confidence in the core functionality without requiring the full multi-GPU benchmarking setup.

Fixes #57.

---

💡 You can make Copilot smarter by setting up custom instructions, customizing its development environment and configuring Model Context Protocol (MCP) servers. Learn more Copilot coding agent tips in the docs.