Add Kronecker discrepancy regression tests (#612)

[algo-hawk]

Summary

Adds unit-test coverage for qmcpy.Kronecker, including regression tests for #612
(periodic_discrepancy returns NaN / is shift-dependent under the default
randomize="SHIFT"). The four tests pinning that bug are marked
@unittest.expectedFailure since the fix has not yet landed on develop
(tracked in PR #556 / PR #633).

• Issue: Related to Kronecker.periodic_discrepancy returns NaN under the default randomize='SHIFT' #612 (does not close it; source fix lands separately)
• Algorithmic or API impact: None — test-only change
• Commands run: pytest test/test_kronecker.py -v (32 passed, 4 xfailed)

Plan for the xfail markers

Once #556/#633 merge #612's fix to develop, I will rebase this branch, remove
the expectedFailure markers from the now-passing tests, and rewrite
test_squared_discrepancy_matches_direct_double_sum (it currently passes only
because it encodes the pre-fix formula's structure).

AI Assistance

• AI assistance substantively affected this PR, and I describe that use below.

Used AI assistance to help design and write the test suite, and to independently
verify (via a scratch worktree) that PR #556 and PR #633 fix the pinned bugs
before recommending them for merge.

Checklist

• I linked the relevant issue or explained why none was needed. -- Related to Kronecker.periodic_discrepancy returns NaN under the default randomize='SHIFT' #612, see plan above.
• I added or updated tests, docs, notebooks, or explained why they were not needed. -- this PR is test-only.
• I verified any equations, citations, benchmarks, numerical claims, or external references changed in this PR. -- reproduced the NaN/shift-dependence bug locally and verified the fix independently against origin/lattice_kronecker and origin/issue_629_choi.
• I reviewed AI-assisted content for licensing, provenance, attribution, confidentiality, and security concerns.
• I described any CI, dependency, notebook runtime, or generated artifact changes if applicable. -- none.