Move cudaIpcMemHandle_t hash and equality to custom namespace

[Copilot]

Addresses feedback to avoid extending std namespace and make template dependencies explicit for the AMD platform handle cache.

Changes:

• Moved std::hash<cudaIpcMemHandle_t> specialization to CudaIpcMemHandleHash functor in anonymous namespace
• Moved global operator== to CudaIpcMemHandleEqual functor in anonymous namespace
• Added noexcept to equality operator (memcmp is non-throwing)
• Updated std::unordered_map to explicitly use custom hash and equality as template parameters

Before:

namespace std {
template <>
struct hash<cudaIpcMemHandle_t> { /* ... */ };
}

inline bool operator==(const cudaIpcMemHandle_t& lhs, const cudaIpcMemHandle_t& rhs) { /* ... */ }

std::unordered_map<cudaIpcMemHandle_t, std::weak_ptr<void>> peerMemoryHandleMap;

After:

namespace {
struct CudaIpcMemHandleHash { /* ... */ };
struct CudaIpcMemHandleEqual { 
  bool operator()(/* ... */) const noexcept { /* ... */ }
};

std::unordered_map<cudaIpcMemHandle_t, std::weak_ptr<void>, 
                   CudaIpcMemHandleHash, CudaIpcMemHandleEqual> peerMemoryHandleMap;
}

---

✨ Let Copilot coding agent set things up for you — coding agent works faster and does higher quality work when set up for your repo.