[fraware]

Post-freeze research/software hardening only. No empirical main changes and no physical-validation claim.

This PR makes repeated-observation research artifacts deterministic under permutation of their input sets:

• canonicalizes observation ordering before incremental Acoustic Object Model clustering;
• makes the model's display specimen ID and source-observation ordering deterministic even when callers vary casing/order;
• uses the canonical model specimen ID in Sonic Twin index entries;
• canonicalizes nuisance characterization pair orientation and ordering;
• rejects duplicate nuisance observation IDs rather than silently producing ambiguous pairs;
• canonicalizes hard-negative generation and adds deterministic tie-breakers;
• canonicalizes benchmark same-specimen and hard-negative pair construction;
• adds regression tests asserting object-model, index, nuisance, hard-negative, and benchmark-pair permutation invariance.

No scoring weights, recurrence thresholds, DSP estimator, evidence schema, empirical Gate semantics, material claim, or calibrated identity claim is changed. The purpose is reproducibility: the same observation set must produce the same research artifacts regardless of caller iteration order.