Use dlopen to load NCCL. (#9796)

This PR adds optional support for loading nccl with `dlopen` as an alternative of compile time linking. This is to address the size bloat issue with the PyPI binary release.
- Add CMake option to load `nccl` at runtime.
- Add an NCCL stub.

After this, `nccl` will be fetched from PyPI when using pip to install XGBoost, either by a user or by `pyproject.toml`. Others who want to link the nccl at compile time can continue to do so without any change.

At the moment, this is Linux only since we only support MNMG on Linux.

python-package/packager/build_config.py

@@ -15,6 +15,8 @@ class BuildConfiguration:  # pylint: disable=R0902
     use_cuda: bool = False
     # Whether to enable NCCL
     use_nccl: bool = False
+    # Whether to load nccl dynamically
+    use_dlopen_nccl: bool = False
     # Whether to enable HDFS
     use_hdfs: bool = False
     # Whether to enable Azure Storage