[bonnhatnguyen]

Description

Adds the first C++ implementation (1.cpp) for the binarytrees benchmark, enabling C++ on the leaderboard.

Key Architectural Highlights

• Thread-Local Reusable Bump Allocator Arena: Instead of incurring thousands of expensive individual heap allocations via new/malloc and individual recursive deallocations, each worker thread retains a pre-sized contiguous memory buffer.
• O(1) Instant Arena Reset: After each tree is constructed and its checksum is evaluated, the arena pointer is reset in $O(1)$ (offset = 0), completely eliminating GC and deallocation overhead.
• OpenMP Guided Parallel Scheduling: Subtrees are distributed across available cores using #pragma omp parallel for reduction(+:chk) schedule(guided) for maximum core utilization.

Verification & Benchmark

• Verified against unit tests:
  • 6 -> 6_out (Match OK)
  • 10 -> 10_out (Match OK)
• Benchmark runtime on depth 18 (18):
  • ~0.17s real time (~1.87s user time across cores), outperforming existing GC and heap-allocated implementations by >15x.