Stabilize cell volumes for highly skew lattice bases

[srijan2607]

Evaluate crystal-cell volume using the equivalent four-sine factorization of the Gram determinant. Compensated degree-space sums reduce cancellation for extreme or near-degenerate cell descriptions; this is not a typical reduced-cell density-corruption claim.

Exact orthogonal results are preserved: cell_volume(2) == 8.0 and cell_volume(2, 3, 4) == 24.0. Flat cells with angles (30, 60, 90), (90, 90, 180) and (120, 120, 120) return exactly 0.0. Real-angle conversion occurs before negation so NumPy unsigned scalars do not wrap.

Verification

The configured default command, python -m pytest, passes 279 tests, including the project's doctests, coverage and mypy checks. Mypy reports no issues in 33 source files; overall coverage is 87%, with periodictable/util.py at 100%. This was rerun on Python 3.13.13, NumPy 2.5.4 and mypy 2.4.0.

The previous head produced 2 failed, 277 passed, with three typing errors in REAL_ANGLE_CASES. The follow-up changes only test-constructor annotations; production code, test configuration and checks are unchanged. Hosted CI has passed: all four test jobs and the wheel build completed successfully; the PyPI upload job was skipped.

No new angle-range validation or general length-product overflow guarantee is added. Known behavior outside the principal angle domain remains outside this repair.

Research basis: NIST DLMF §4.21 supplies the addition/product identities; the IUCr coreCIF cell-volume definition supplies the Gram determinant. Floating-point accuracy is tested separately from the algebraic identity.

[srijan2607]

The test-typing follow-up is pushed at 0dbbf7d0a4d2. The configured local python -m pytest run passes 279 tests, including the mypy checks.

@pkienzle could you approve the pending fork workflow when appropriate? The current-head run is waiting for approval with zero jobs executed; hosted CI is not yet verified.

• https://github.com/python-periodictable/periodictable/actions/runs/38904881548