linalg: add kron

[axiom-of-choice]

Adds kron(A, B) to aneforge.linalg, built as a broadcast-multiply of expanded views plus a reshape (no new kernel):

• A.expand_dims((1, 3)) * B.expand_dims((0, 2)) → [m, p, n, q], then reshape to [m*p, n*q].

Closes #126.

Testing

• test_kron in tests/test_linalg.py (two non-square shapes) compares against np.kron; on-device M2 Pro: passed (max abs err 4.9e-4, ~half ULP fp16).
• Off-device suite: 407 passed, 2 skipped.
• ruff, pylint 2-space gate, pyright aneforge, compileall clean.

[sbryngelson]

Verified on device (M5): the kron test passes, and the full linalg suite is unaffected.

The expand_dims/broadcast-multiply/reshape construction is the right shape for this, and
_check_precision=False matches the existing convention in this module (linalg.py:36, 360, 716).
Nothing to change.

One scheduling note: #259, #260 and this all append to the same __all__ block and the same tail
of linalg.py, so whichever lands first will force a rebase on the other two.