Fix #89: make THINK vs SAY divergence observable

[RandomOscillations]

Summary

• add public_position to ReasoningResponse so high-fidelity reasoning can carry a separate public stance
• classify private position from private_thought (fallback to reasoning), and classify public position from public_statement at high fidelity
• apply response.public_position in engine state updates (fallback to response.position) so divergence persists to DB-visible state
• include fidelity in meta.json for run auditability
• add tests for async two-pass divergence classification, engine public-position override behavior, and fidelity metadata export

Testing

• pytest -q tests/test_reasoning_execution.py tests/test_engine.py::TestFlipResistance::test_public_position_prefers_explicit_public_field tests/test_engine.py::TestTokenAccumulation::test_cost_in_meta_json
• ruff check extropy/core/models/simulation.py extropy/simulation/engine.py extropy/simulation/reasoning.py tests/test_engine.py tests/test_reasoning_execution.py

Closes #89

[DeveshParagiri]

Code Review

Verdict: ✅ Ready to merge

Summary

Makes THINK vs SAY divergence observable by classifying private_thought and public_statement separately in high-fidelity simulations.

Implementation

Component	Change
ReasoningResponse	Added public_position: str | None
Classification	Separate calls for private_thought (→position) and public_statement (→public_position)
Engine	Uses response.public_position or response.position for DB state
meta.json	Now includes fidelity for auditability

Edge Cases

Case	Handling
Empty public_statement	Returns {}, falls back to position
No divergence	public_position equals position
Medium/low fidelity	Skips public classification

Minor Suggestion (non-blocking)

Consider adding a test for empty public_statement edge case to verify fallback works.

No blocking changes required.