test(hrv): split flaky integration into synthetic + floor + opt-in real-data

[neuron7x]

Summary

Replaces #168 (margin bandaid n_subjects=3→5) with a first-principles split.

The previous TestHRV.test_gamma_in_range welded three orthogonal claims onto one mandatory mid-CI network call:

1. compute_gamma mathematics recovers γ ≈ 1 from a 1/f PSD,
2. _ensure_loaded enforces the 3-subject floor,
3. real PhysioNet/NSR2DB data lies in the physiological band [0.5, 1.5].

When wfdb/PhysioNet flakes — as happened in run 25184505044 — claims (1) and (2) become non-testable. Bumping n_subjects only widens the absorption margin; the substrate is still the network.

Decomposition

Test	Always-on	Network	Catches
test_compute_gamma_recovers_unity_from_synthetic_1f	✅	❌	γ-estimator math regressing on 1/f PSD
test_floor_minimum_enforced	✅	❌ (monkeypatched _load)	floor=3 enforcement breaking
test_real_data_gamma_in_range	opt-in (NEOSYNAPTEX_RUN_NETWORK_TESTS=1)	✅	full wfdb→PSD→γ on real NSR2DB

The floor constant is named (_HRV_FLOOR_MIN_SUBJECTS = 3), not magic. The CI-run-id comment from #168 is removed: the test no longer depends on transient network state, so diagnostic context lives in the commit message rather than polluting source.

Effect on CI

• PR-blocking matrix (Verify (3.10/3.11/3.12)) becomes fully offline for the HRV substrate. wfdb/PhysioNet flake can no longer break PRs.
• Real-data integration is preserved as an opt-in gate; running locally:

  NEOSYNAPTEX_RUN_NETWORK_TESTS=1 pytest tests/test_integrity_v2.py::TestHRV::test_real_data_gamma_in_range
  

• Follow-up suggested (out of this PR's scope): a scheduled CI lane that sets NEOSYNAPTEX_RUN_NETWORK_TESTS=1 nightly so PhysioNet API drift is still caught.

Test plan

• pytest tests/test_integrity_v2.py → 16 passed, 1 skipped (real-data, as designed).
• ruff format + ruff check clean.
• Synthetic test recovers γ within [0.7, 1.3] (deterministic seed 42).
• Floor test asserts the documented RuntimeError("Insufficient HRV data").
• Local pre-commit hooks pass.

Claim status

claim_status: derived

Pure test-architecture refactor — zero new measurement claims. The synthetic test recovers a textbook 1/f→γ=1 relation; the floor test exercises documented error handling; the real-data test is unchanged in intent, only gated.

Supersedes #168.

🤖 Generated with Claude Code