Port the structured idle-noise families to the native-surface NoiseModel (twirl/threshold route)

[ciaranra]

Goal

Port the structured idle-noise interface (PR #420: p_idle_linear/p_idle_linear_model,
p_idle_sin_squared/p_idle_sin_squared_model, p_idle_coherent/p_idle_coherent_model,
relative-rate multiplier semantics, symmetric defaults, first-class L key) to the
native-surface noise surface used by the twirl and threshold pipelines, so both routes share
one vocabulary and one set of fail-loud rules.

Current state

• pecos.qec.surface.decode.NoiseModel exposes only the legacy idle fields (p_idle,
  T1/T2, per-axis rates, bare Z-only aliases) and forwards them directly to the Rust
  DemBuilder.with_noise (python/quantum-pecos/src/pecos/qec/surface/decode.py).
• The Pauli-twirl handoff route builds DEMs exclusively through that surface, so it never
  sees the structured families, the ambiguity/clobber conflicts, or the deprecation
  warnings for the bare aliases (tests/qec/surface/test_pauli_twirl_handoff.py exercised
  them until the alias migration on PR Align idle-noise vocabulary across DEM, engines and neo, and emit the requested Pauli channel #420's branch).
• generate_dem_from_tick_circuit and the pure-Python Pauli-frame DEM builder are likewise
  legacy-only surfaces (python/quantum-pecos/src/pecos/qec/surface/circuit_builder.py).

Work

• Reuse the shared translator from pecos/qec/dem.py (_translate_structured_idle_noise
  and the family/model validator) rather than duplicating semantics; one implementation,
  two consumers.
• Same conflict, deprecation, and rejection behavior as the Guppy entry points, including
  rejecting nonzero coherent rates on the standard DEM route.
• Decide the fate of NoiseModel.p_idle consistently with the Guppy surface (the shorthand
  was removed there in favor of explicit p_idle_linear).

Consistency principle

This is one step of a broader goal: identical idle-noise semantics at every level that
accepts noise parameters, including below the public API (the Rust from_circuit binding's
idle_rz currently accepts a coherent angle and silently stores its Pauli twirl, while the
Guppy surface rejects requested coherence loudly — that split should eventually close in
the same direction).

[ciaranra]

Done in PR #420 (commit b04a10da9).

The translator and family validator moved out of pecos/qec/dem.py into a shared private module pecos/qec/_idle_noise.py, and NoiseModel (pecos/qec/surface/decode.py) now accepts the same three families. One implementation, two routes — the Guppy entry points and the native-surface/twirl route — so conflicts, deprecation warnings for the bare Z-only aliases, model validation, and the "L"-key rules are shared rather than reimplemented.

Design point worth recording: NoiseModel.__post_init__ translates the families into the canonical per-axis fields and then clears the family fields. That is load-bearing, not tidiness: for_runtime_idle_time_units() uses dataclasses.replace, which re-runs __post_init__, so if the family fields persisted alongside their translated outputs the second pass would see structured-plus-primitive and raise the conflict error. A round-trip test covers exactly that.

No idle_rz field was added to NoiseModel. The standard DEM route rejects nonzero coherent idle noise outright, so a field there would only ever hold a value that cannot be used; zero-valued coherent input validates and clears.

Verification: the pre-existing Guppy suites are the regression oracle for the extraction and are unchanged (306 passed, 1 xfailed across test_from_guppy_dem.py, test_guppy_dem_build.py, the new test_idle_noise_families.py, and test_pauli_twirl_handoff.py); repo-wide lint clean on a cold run with the ruff cache cleared.

Remaining for the wider vocabulary work: the five Rust signatures (from_circuit, DemBuilder.with_noise, DemSampler.from_circuit, DemSampler.with_detectors, DemSamplerBuilder.with_noise) still take the flat parameter block. This issue covered the Python-side fork, which was the one that had two competing vocabularies.