Add two-point Kronecker multiplication

[kim-em]

Implements the first coefficient-owner kernel from the multipoint Kronecker milestone in SPEC/Libraries/hex-poly-fast.md.

• adds forced Hex.ZPoly.mulKronecker2, evaluating at B and -B with exactly two packed Int multiplications
• derives the slot bound from min(size) * maxAbs(p) * maxAbs(q)
• proves signed digit recovery, even/odd stream separation and interleaving, and mulKronecker2_eq
• extends ZPoly conformance with signed, reversed-operand, and digit-boundary KS2 cases
• extends the existing crossover grid with KS2 timing and checksum agreement

The full local crossover sweep found KS2 faster than KS1 in only 2/276 cells on this machine. This PR therefore intentionally exposes KS2 as a forced kernel without putting it in production dispatch; the measured dispatcher remains work for the completed KS2/KS3/KS4 comparison.

Validation:

• lake build HexPolyZ HexPolyZ.Conformance HexPolyZ.KroneckerCrossover
• full 276-cell hexpolyz_kronecker_crossover sweep with matching checksums
• python3 scripts/check_phase4.py --base origin/main
• git diff --check