[Dev] Preserve explicit NCCL NVLS configuration

[lhb8125]

Summary

• default NCCL_NVLS_ENABLE to 1 only when the caller has not set it
• preserve an explicit NCCL_NVLS_ENABLE=0 or 1 from the launch environment
• cover both the default and explicit-override behavior with a unit test

This change is not required when a recipe always runs with NVLS enabled, but it is required for valid NVLS A/B experiments and for preserving caller-controlled NCCL semantics.

Test plan

• CHECK_ONLY=true BASE_REF=dev uv run --no-sync --project /home/scratch.hongbinl_sw/work/fsdp/agentic-mcore-dev --extra mcore-lint -- bash tools/autoformat.sh
• python tools/check_copyright.py megatron/core/distributed/distributed_data_parallel.py tests/unit_tests/distributed/test_distributed_data_parallel.py

The local toolkit environment does not contain PyTorch, so import-dependent pytest/mypy execution is deferred to the Megatron CI container.

🤖 Generated with Codex

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.