[EM] More rigorous check for potential cross socket access.

[trivialfis]

• Move the get cpu function.
• Use has_normal_memory and has_cpu to avoid false positives.
• Check invalid node bind.
• Workaround old CTK for the host cache.
• Cleanup version functions.

[trivialfis]

cc @rongou .

[Copilot]

Pull Request Overview

This PR implements more rigorous checks for potential cross-socket memory access in NUMA systems, focusing on reducing false positives and improving validation. The changes enhance NUMA topology detection by adding functions to check for CPU and normal memory nodes, refactor version checking utilities, and improve error handling for host cache operations.

• Move CPU NUMA detection functions from threading utilities to NUMA topology module
• Add rigorous NUMA validation using has_normal_memory and has_cpu system files
• Refactor and consolidate CUDA version checking functions with clearer naming

Reviewed Changes

Copilot reviewed 15 out of 15 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
tests/cpp/data/test_sparse_page_dmatrix.cc	Add filesystem header and improve include comments
tests/cpp/common/test_numa_topo.cc	Add tests for new NUMA CPU and memory node detection functions
tests/cpp/common/test_device_vector.cu	Update function call to use renamed version checking function
src/data/ellpack_page_source.h	Remove NUMA include and add declaration for extracted NUMA check function
src/data/ellpack_page_source.cu	Implement rigorous NUMA checking logic and add driver version validation for host cache
src/common/threading_utils.h	Remove CPU NUMA function declaration (moved to numa_topo.h)
src/common/threading_utils.cc	Remove CPU NUMA function implementation (moved to numa_topo.cc)
src/common/numa_topo.h	Add new NUMA detection functions and CPU NUMA function from threading utils
src/common/numa_topo.cc	Implement new NUMA detection functions and moved CPU NUMA function
src/common/device_helpers.cu	Fix namespace prefix for version checking function
src/common/device_compression.cu	Add null pointer check for memory pool
src/common/cuda_rt_utils.h	Rename version functions with clearer "Global" suffix
src/common/cuda_rt_utils.cc	Implement renamed version functions
src/common/cuda_dr_utils.h	Remove duplicate version function declaration
src/common/cuda_dr_utils.cc	Update to use renamed version function and remove duplicate implementation