Use TensorCore tensor interface

[Krastanov-agent]

Summary

• extend the shared TensorCore tensor and ⊗ bindings
• preserve the existing QuantumInterface tensor methods and zero-argument error
• add binding regression coverage and prepare version 0.4.4

Tests

• focused basis tests
• Aqua
• full QuantumInterface suite (running)
• dependent package JET/Aqua suites (running)

[codecov]

Codecov Report

❌ Patch coverage is 33.33333% with 2 lines in your changes missing coverage. Please review.
✅ Project coverage is 34.22%. Comparing base (adda232) to head (8cd0c14).
⚠️ Report is 1 commits behind head on main.

Files with missing lines	Patch %	Lines
src/tensor.jl	0.00%	2 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main      #93      +/-   ##
==========================================
+ Coverage   34.14%   34.22%   +0.07%     
==========================================
  Files          15       15              
  Lines         451      450       -1     
==========================================
  Hits          154      154              
+ Misses        297      296       -1     

☔ 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.

[Krastanov-agent]

Validation update:

• The full QuantumInterface suite passes locally on Julia 1.12.6; PR CI passes Julia 1.6, current Julia, and Julia alpha.
• The JET job fails before analyzing QuantumInterface because the existing test imports the removed JET.ReportPass API from JET 0.12.1.
• This is a pre-existing baseline failure: it reproduces on untouched upstream/main locally and in the latest main-branch JET job.

I am leaving the unrelated JET harness migration out of this minimal PR. Downstream Aqua/JET validation is still running.

[Krastanov-agent]

Downstream validation completed on Julia 1.12.6 with local QuantumInterface 6b9aea1 (v0.4.4) and TensorCore 0.1.1:

Package	Aqua/general	JET	Result
QuantumOpticsBase	2,270 pass, 2 expected broken	86 pass, 69 marked broken	pass
QuantumClifford	346,547 pass, 9 expected broken	39 pass, 2 expected broken	pass; direct tensor(bell_state, bell_state) JET check passes
QuantumSymbolics	Aqua 12/12; general 481 pass, 2 expected broken	1 pass, 1 expected broken	pass
Gabs	1,245 pass, 4 expected broken	pass at existing 15-report threshold	pass; shared tensor/⊗ identity probe passes
QuantumSavory	Aqua 12/12	one report/failure	identical report with registered QI 0.4.3; baseline, unrelated to tensor
QuantumOptics	Aqua passes; 2,667 pass, 1 expected broken, 2 allocation assertions fail	43/43	allocation failures reproduce with identical counts on QI 0.4.3; baseline

No downstream source changes were needed. An independent review of both commits found no issues. Hosted QOB and QuantumSymbolics breakage jobs are green; the remaining hosted jobs are still completing.

[Krastanov-agent]

Hosted downstream breakage run is complete:

• PASS: QuantumOpticsBase, QuantumSymbolics, and QuantumClifford.
• QuantumOptics failed on the same two allocation-equality assertions reproduced locally with both this branch and released QuantumInterface 0.4.3. Its hosted Aqua failure is caused by the breakage workflow adding QuantumInterface as a direct dependency; the same check was already red on Specialize bipartite basis transformations for inference #92.
• QuantumSavory passed 15,120 tests and failed only Aqua's persistent-task probe after its precompilation subprocess did not create done.log. The same hosted Aqua check was already red on Specialize bipartite basis transformations for inference #92; local Aqua passed 12/12. Its separate local JET finding was also unchanged with released QuantumInterface 0.4.3.

Run: https://github.com/qojulia/QuantumInterface.jl/actions/runs/33414496806

No downstream failure attributable to the TensorCore migration was found. I have not changed unrelated downstream baselines.

[Krastanov-agent]

Follow-up pushed in e84e7e0:

• removed the zero-argument tensor() extension;
• made the Basis, AbstractOperator, and StateVector variadic reducers require a first owned argument, avoiding the three zero-length vararg ambiguities;
• restored unconfigured Aqua.test_all(QuantumInterface).

Focused results on Julia 1.12.6:

• basis tests: 38/38 pass;
• Aqua: 11/11 pass, including piracy and ambiguity checks;
• full non-JET suite: 73/73 pass;
• explicit dispatch check: no zero-argument applicability, no tensor ambiguities, and the function remains owned by TensorCore.

Downstream Aqua/JET checks are running against this commit.

[Krastanov-agent]

Local downstream follow-up for e84e7e0 is complete:

• QuantumOpticsBase: tensor behavior and JET pass. Aqua exposes one new zero-arity ambiguity between tensor(Ket...) and tensor(Bra...); the broader suite has 2,269 pass, 1 Aqua failure, and 2 expected broken.
• QuantumOptics: tensor behavior, Aqua, and JET all pass.
• QuantumClifford: tensor behavior and JET pass at the prior baseline. Aqua exposes 10 zero-arity intersections among its existing typed-vararg tensor methods; every other Aqua category, including piracy, passes.
• Gabs: tensor behavior and its declared JET baseline pass. It has no declared Aqua test; a supplemental check finds no tensor piracy.
• QuantumSymbolics: JET remains at the prior accepted baseline. Aqua is 11/12 because its existing tensor(xs::SymQObj{T}...) now exposes an unbound T at zero arguments.
• QuantumSavory: Aqua passes 12/12. JET has exactly the same sole switches.jl:174 report previously reproduced with QuantumInterface 0.4.3.

All concrete tensor identity/operation checks pass and all packages resolve the same TensorCore-owned generic. Removing QuantumInterface's global tensor() disambiguator necessarily exposes downstream methods whose typed varargs still accept zero arguments. The unobtrusive downstream remedy is the same leading-owned-argument signature used here; I have not edited those repositories.

[Krastanov-agent]

Hosted breakage run is complete: https://github.com/qojulia/QuantumInterface.jl/actions/runs/33423155538

• QuantumOpticsBase confirms the local result exactly: 2,269 pass, one new Ket/Bra zero-arity Aqua ambiguity, 2 expected broken.
• QuantumClifford confirms the local result: 359,495 pass, one Aqua failure containing the 10 zero-arity tensor ambiguities, 9 expected broken.
• QuantumOptics has only its previously reproduced two allocation assertions plus the breakage workflow's direct-dependency Aqua artifact.
• QuantumSavory again passes 15,120 tests and fails only Aqua's hosted persistent-task subprocess probe, as on the preceding run.
• The QuantumSymbolics hosted runner lost communication with GitHub before publishing logs. Its local declared checks are complete: JET unchanged; Aqua 11/12 on the existing zero-argument-capable symbolic tensor method's unbound T.

No additional source change is indicated. Downstream zero-length vararg signatures need repository-local follow-ups; restoring QuantumInterface.tensor() would reintroduce the piracy this update removes.