test: witness coverage for 4 under-tested invariants (#222)

* test: witness coverage for 4 under-tested kuramoto/physics modules

Adds INV-* anchored witness tests for modules that were either at 0%
coverage (core/kuramoto/falsification, core/kuramoto/metrics) or had
behavioural tests without physics-contract anchors (core/physics/landauer,
core/physics/tsallis_gate).

New test files (all physics-validator L1-L5 clean):

* test_T24_kuramoto_metrics_witness.py — 8 tests
    - INV-K1 bound on order_parameter via Hypothesis + antipodal witness
    - INV-SB2 determinism of permutation_entropy
    - INV-HPC2 chimera_index on disconnected graph; rolling_csd window edges

* test_T25_falsification_surrogates.py — 8 tests
    - INV-HPC1 seeded replay of IAAFT surrogate + different-seed falsifier
    - INV-HPC2 amplitude multiset preservation + spectrum residual bound
    - INV-K1 bound on time_shuffle_test observed + null samples
    - INV-HPC2 degree-preserving rewire preserves degree sequence

* test_T26_landauer_witness.py — 9 tests
    - INV-TH2 non-negative entropy per step (property + falsifier)
    - INV-HPC2 formula equality to k_B·T·ln2, ratio finiteness, temp rejection
    - INV-HPC1 bit-identical replay across runs

* test_T27_tsallis_gate_witness.py — 11 tests
    - INV-FE2 position_multiplier ≥ 0 via Hypothesis + crisis-floor falsifier
    - INV-HPC2 estimate_q finite and ≥ 1; zero-variance guard
    - INV-SB2 deterministic replay across fresh instances

Targeted coverage delta (before → after, using .coveragerc source tree):

    core/kuramoto/falsification.py : 0.00%  → 54.14%  (+54.1 pp)
    core/kuramoto/metrics.py       : 0.00%  → 40.99%  (+41.0 pp)
    core/physics/landauer.py       : 94.74% → 96.49%  (+1.75 pp)
    core/physics/tsallis_gate.py   : 90.43% → 96.81%  (+6.38 pp)

Every tolerance in the new tests is derived from a documented formula
(float64 unit roundoff, IAAFT convergence bound, Bickel-Freedman kurtosis
std error, Molloy-Reed degree-sequence invariant). Every new test
includes a falsification probe per the ladder discipline.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

* style: apply black to new T24/T26 witness tests

---------

Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Co-authored-by: Yaroslav Vasylenko <neuron7x@users.noreply.github.com>

Author: 3 people