Decode Laya answers on CPU

[linear3735]

Superseded by #90, now merged into main.

Purpose

Decode raw option and action logits into Laya answers, probabilities and confidence. Preserve question and option order and apply temperature calibration.

Depends on #43, which contains the preprocessing layer. #21 is merged. The current diff includes #43. Original decoder increment contains 159 core/configuration lines; tests and fixtures are separate. Scorer execution is separate. Refs #14.

This branch inherits #43's private-key JSON fix, explicit Request::from_json / Request::from_value API and root test relocation. Generic Deserialize is removed.

Test Plan

• Run workspace tests, fmt, Clippy and release build.
• Regenerate the reference with Laya 0.3.20 and compare it byte-for-byte.
• Check malformed rows, metadata, non-finite logits and temperature boundaries. Commands and fixture regeneration are in the CPU decoder recipe.

System1-Omni Version / Commit: bc90ffc. Original decoding increment: 99c1d12 → d9e57b7.

Test Result

• Local workspace and current CPU CI: 45 tests passed and 6 external-data/GPU tests were skipped. The inherited 17-case packing check passed separately.
• The checked-in 16-case official reference and the FP32 rounding probe passed. The probe allows a probability difference of 0.0001; all other fields match exactly.
• fmt, strict Clippy, release build and strict MkDocs build passed locally.
• Earlier fixture regeneration was byte-identical. Decoder code and the fixture are unchanged; no GPU inference or model-quality checks were run for this update.

CI for bc90ffc: Rust CI, Docs build, benchmark harness tests passed.

Self-review

Before marking this PR ready for review or requesting maintainer review, complete
the self-review checklist.
Keep the PR in draft while this work is incomplete.
For agent assistance, use the optional precheck-pr skill.

• I have reviewed the full diff and addressed the issues I found.
• I have checked that the change follows the project's architecture and stays focused on the stated purpose.
• I have run the checks appropriate to this change and reported commands, results, and anything I could not verify above.
• I have checked that the PR description, documentation, and any accuracy or performance claims match the implementation and available evidence.

[hsliuustc0106]

Independent local review — Laya answer decoding

Verdict: approved.

decision.rs completes the processor pair: raw option/action logits → answers, probabilities, confidence with temperature calibration and preserved question/option order; the export_decisions.py oracle and tests/laya/data/decisions.json fixtures keep it reference-checked, and postprocessing stays separate from scorer execution exactly as docs/architecture.md draws the line. Locally at head bc90ffcf (includes #43): laya suite 25 passed / 3 ignored (checkpoint-gated), clippy clean; CI green. Same json.rs rebase note as #43. Merge after #43.

Approval per the repo review process; reflects head bc90ffcf only.

[hsliuustc0106]

fix conflicts as the new arch