refactor(core): deepen internal execution seams

[LessUp]

Summary

• internalize texture preparation behind a non-public seam and add an architecture test locking it down
• move CSR/ELL device lifecycle code into dedicated internal modules and keep format files focused on host logic
• add shared PageRank helpers plus a working CPU-only backend for no-CUDA builds

Test Plan

• cmake -S . -B build-no-cuda -DSPMV_REQUIRE_CUDA=OFF
• cmake --build build-no-cuda -j$(nproc)
• ctest --test-dir build-no-cuda --output-on-failure