Symmetric memory pytorch backends

[saivishal1999]

Adds support for PyTorch-backed symmetric memory in nvFuser alongside the existing native implementation. This enables using c10d::symmetric_memory (NCCL/NVSHMEM/CUDA) as an alternative backend while keeping the current SymmetricTensor API unchanged.

• Introduced SymmetricMemoryBackend (Native, PyTorchNccl, PyTorchNvshmem, PyTorchCuda) with runtime selection.
  --Default behavior unchanged (native backend).
  --PyTorch backend is opt-in via NVFUSER_ENABLE=symmetric_memory_backend(...).
• Integrated PyTorch symmetric memory into SymmetricTensor:
  allocate (empty_strided_p2p)
  setupRemoteHandles (c10d::symmetric_memory::rendezvous)
  remoteTensor(get_remote_tensor)
• Added process group tracking + registration in Communicator to support rendezvous and ensure proper cleanup.
• Extended c10d mocks to allow builds without distributed support.