Fix Jasp quantum-quantum QuantumModulus multiplication in QuantumArray

[alighazi288]

Description

Fixes element-wise quantum-quantum multiplication of QuantumArrays of QuantumModulus in dynamic (Jasp) mode, which returned a value scaled by a power of two. e.g. 1 * 5 mod 7 gave 3 instead of 5.

Root cause: qq_montgomery_multiply_modulus computed the Montgomery reduction shift as smallest_power_of_two((N-1)**2 + 1) - n, whereas the static montgomery_mod_mul and create_output_qf use ceil(log2((N-1)**2) + 1) - n. Since ceil(log2(k²+1)) ≠ ceil(log2(k²)) + 1, the jasp result's res.m came out one too small for many moduli (7, 11, 13, …). A single QuantumModulus product returns that result directly and still decodes correctly, but the QuantumArray element-wise path preallocates the output's Montgomery shift via create_output_qf, so the injected result was decoded against a mismatching shift and came out off by 2¹. The fix aligns the jasp shift with the static formula (JAX/BigInteger-safe — the + 1 moves outside smallest_power_of_two).

Related Issues

Closes #640

Type of Change

• Bug Fix

Breaking Change?

• No

What was changed?

• Corrected the Montgomery reduction shift in qq_montgomery_multiply_modulus (smallest_power_of_two((N-1)**2 + 1) - n → smallest_power_of_two((N-1)**2) + 1 - n) so it matches the static montgomery_mod_mul / create_output_qf convention, with a comment documenting the invariant.
• Removed the pytest.skip guarding the QuantumModulus quantum-quantum cases in test_quantum_array_element_wise_ops_qm, so they now run.

How was it tested?

Test-ID	Description	Status
T-001 (Positive)	a * b element-wise on QuantumArray[QuantumModulus] returns (A*B) % N - issue repro (1*5 mod 7 = 5) + random 2×2/3×2 arrays over N ∈ {7,11,13,15,23,31,47}	✅
T-002 (Negative)	Quantum-classical mul on a QuantumArray of non-QuantumModulus still raises NotImplementedError (unchanged)	✅
T-003 (Boundary)	Standard-form (m=0) and pre-shifted (m≠0) single-QuantumModulus products remain correct; smallest modulus exercised via scalar-RHS quantum case (N=3)	✅
T-004 (Boundary)	Chained (a*b)*c and non-square arrays across several moduli	✅

Local suites green: test_jasp_quantum_array.py (113), test_jasp_modulus.py (56, incl. the qq standard-form / non-standard-m / traced-BigInteger regressions), test_montgomery.py. Also confirmed the corrected shift equals the static formula for every odd modulus 3–63.

Checklist

• My code follows the project's coding standards
• I have performed a self-review
• I have added/updated tests (referencing issue Test-IDs)
• All tests pass locally and in CI
• I have updated the documentation
• I have added a changelog entry
• Breaking changes are documented with migration path

Reviewer Notes

The previously-skipped test_quantum_array_element_wise_ops_qm (mul, quantum RHS) is now un-skipped and serves as the regression guard (its N=7 array case reproduces the bug).

Design note: this aligns the jasp shift with the existing canonical formula, which makes it one larger for affected moduli so qq_montgomery_multiply_modulus uses ~1 extra ancilla qubit, matching what the static path already allocates. I chose this fix over the qubit-optimal alternative (keep the tighter jasp shift and instead have the injection carry the actual res.m), which would touch the well-tested create_output_qf/injection machinery. Happy to go the other route if you'd prefer.