Drop un-combined zero coefficients when collecting like terms

[oameye]

_addto_key! collects like terms and deletes a key when its combined coefficient is zero. It used the structural _iszero_cnum plus an exact-negation fast path (added in #162 for γ/D + (-γ)/D). Both miss cancellations whose two halves are equal but differently represented rationals, e.g. λ/2 vs (1//2)λ. Such a term is mathematically zero but survives in the QAdd.

Concretely, for a quadratic (squeezing) Hamiltonian H = Δ·a†a + (λ/2)(a†² + a²):

commutator(im*H, a*a)
# kept:  ... + (im*(λ/2 - (1//2)λ)) * a†a†aa + (im*(λ/2 - (1//2)λ)) * aaaa
#        i.e. two spurious 4th-order operators with a zero-but-un-combined coefficient

iszero on that coefficient returns true (Symbolics expands), and average drops it, but the operator-level QAdd retained it. Downstream consumers (e.g. QuantumCumulants' cumulant/average_and_truncate path) then had to special-case the phantom terms, and an exact (no-truncation) mean-field expansion of a quadratic Hamiltonian failed to close because completion chased these never-vanishing higher-order moments.

Fix

Add _iszero_cnum_expand (an expanding zero test) and use it as a final fallback in _addto_key!, gated behind the existing _is_symbolic_cnum guard:

• numeric and structural fast paths are unchanged (common merges stay allocation-free);
• the expand runs only for symbolic coefficients that did not cancel structurally or by exact negation, which is exactly the problematic case.

After the fix, commutator(im*H, a*a) returns only the 0th/2nd-order terms, and (λ/2)·aa - (1//2)λ·aa collapses to an empty (iszero) QAdd.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 94.48%. Comparing base (08fb182) to head (3f78501).

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #167      +/-   ##
==========================================
+ Coverage   94.44%   94.48%   +0.03%     
==========================================
  Files          23       23              
  Lines        2090     2105      +15     
==========================================
+ Hits         1974     1989      +15     
  Misses        116      116              

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[Copilot]

Pull request overview

Fixes a symbolic-prefactor cancellation edge case in like-term collection so mathematically zero terms don’t remain in QAdd when the two cancelling coefficients are equal but represented with different Symbolics expression trees (e.g. λ/2 vs (1//2)*λ), which previously caused spurious higher-order operator terms to persist after merges (notably in commutators).

Changes:

• Improve the exact-negation cancellation check by canonicalizing p / n (integer-literal division) to (1//n) * p before isequal comparison.
• Add a regression test covering the λ/2 vs (1//2)λ commutator scenario to ensure no spurious 4th-order terms survive.
• Document the fix in the changelog under Unreleased.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 2 comments.

File	Description
test/arithmetics/commutator_test.jl	Adds a regression test ensuring differently-represented zero coefficients cancel and don’t leave higher-order operator terms.
src/expressions/cnum.jl	Updates symbolic negation detection with a lightweight canonicalization for division-by-integer cases to enable cancellation.
Changelog.md	Adds an Unreleased entry describing the cancellation fix and its impact.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.