ci: GitHub Actions workflow + base pipeline unit tests

[PlayerPlanet]

Sets up continuous integration via GitHub Actions and adds comprehensive unit tests for the base QEC pipeline. This gives us automatic test runs and coverage reporting on every push and pull request, so we can catch breakage before merge.

What

.github/workflows/ci.yml — GitHub Actions CI

• Triggers on push to main and every pull_request to main
• Python 3.13 on ubuntu-latest
• Two jobs:
  1. test: installs runtime + test deps, runs pytest with coverage, fails if under 85%
     • Generates term, XML, and HTML coverage reports
     • Uploads to Codecov (optional, fails open)
  2. smoke: runs python -m diffqec.smoke_test end-to-end
• 20-minute timeout per job
• Pip cache keyed on requirements.txt and pyproject.toml

tests/test_pipeline.py — 49 new unit tests

Organized by module:

• TestMakeStimCircuit (8 tests): default Z memory, X memory, case-insensitive memory arg, no_reset flag, noise model application, partial noise keys, distance 5
• TestMakeQiskitCircuit (9 tests): return type, metadata keys, distance 2/3 counts, X/Z memory, no_reset, depth scaling, noise param ignored
• TestStimToQiskit (5 tests): basic conversion, dense contiguous indices, strips noise instructions, measurement order length, raises on unhandled gates
• TestGetCircuitInfo (3 tests): expected keys, qubit classification disjoint, distance-3 counts
• TestGetQubitLists (2 tests): data/anc split, no-coords fallback
• TestGetMeasOrder (1 test): order matches circuit
• TestCountsToMeasurementArray (3 tests): basic conversion, explicit count, truncation
• TestExtractSyndromesManual (6 tests): noiseless zero detections, output dict keys, shape, error injection, noise model, print summary
• TestExtractSyndromesStim (3 tests): detector sampler, noiseless zero, extract_syndromes with raw_meas
• TestBuildEmeraldQubitMap (2 tests): returns dict, import contract
• TestDecodeHardwareResults (3 tests): mwpm default, diffqec missing model_path raises, unknown decoder
• TestMainSimulation (2 tests): run_simulation_test runs, main entrypoint works
• TestPipelineIntegration (2 tests): circuit info consistent, stim→qiskit simulation

tests/__init__.py

Empty package marker for tests/.

pytest.ini

• testpaths = tests
• Coverage source: surface_code, extract_syndromes, internal_helpers, run_on_hardware, diffqec
• Omits: tests/, iqm_utils/, benchmarks/, lumi/, main.py, build_emerald_qubit_rotated.py
• Filter warnings for qiskit/stim/pymatching deprecations
• slow and hardware markers for selective deselection

requirements.txt

• Added pytest>=8.0
• Added pytest-cov>=5.0

Test results

• 64 tests passing (49 new + 15 existing diffqec tests)
• 86% line coverage (CI threshold: 85%)
• ~5 seconds total runtime in CI

Coverage report

Module	Coverage
extract_syndromes.py	100%
internal_helpers.py	100%
diffqec/denoiser.py	100%
diffqec/diffusion.py	100%
diffqec/syndrome_processor.py	100%
surface_code.py	90%
diffqec/model.py	95%
diffqec/data.py	80%
diffqec/integrate.py	81%
run_on_hardware.py	44% (hardware path, not testable in CI)
main.py	76% (print path)

Next steps

When this is merged, every future PR (including the LUMI benchmark) will automatically run the full test suite and report coverage, catching regressions before they reach main.