[FEA]: Support for finding nvshmem host library via cuda-pathfinder #823

Is this a duplicate?

• I confirmed there appear to be no duplicate issues for this request and that I agree to the Code of Conduct

Area

cuda.pathfinder

Is your feature request related to a problem? Please describe.

Currently we depend on customlly defining LD_LIBRARY_PATH / LD_PRELOAD to determine the path to libnvshmem to support nvshmem4py features. We should delegate the finder to cuda-pathfinder.

Third party developers like deepEP implement their own logic: https://github.com/deepseek-ai/DeepEP/blob/695b6347b4f3c6e4906f11e8b8dce0d1e2058454/setup.py#L10-L31 and cuda-pathfinder should relieve them of this responsibility.

Note this isn't a duplicate of #661. This PR finds the headers to nvshmem. The searched library priority should stay aligned with the searched header path.

Describe the solution you'd like

One solution for the consistency problem is to use relative path from the searched library to the headers.

Describe alternatives you've considered

No response

Additional context

No response