Speed up point-subgroup traversal

[lan496]

Summary

• replace the thread-safe Queue in point-subgroup traversal with collections.deque
• mark group elements visited when enqueued to avoid duplicate traversal work while preserving result order
• compare the complete ordered subgroup result with brute-force enumeration in the regression test

Related: https://github.com/lan496/spinforge-dev/issues/193

Test plan

• .venv/bin/python -m pytest -q (136 passed, 1 skipped)
• tests/configuration_tests/test_oriented.py::test_noncoplanar_UO2_index4 against local spgrep
• compare UO2 runtime with the old and new traversal (24.62 s -> 16.18 s)
• commit-time formatter, Ruff, and mypy hooks on the changed files

Notes

prek run --all-files still reports unrelated pre-existing Ruff findings elsewhere in the repository; the hooks for the files changed by this PR pass.