Fix #87: enforce intent→behavior accountability in re-reasoning

[RandomOscillations]

Summary

• strengthen re-reasoning prompt with explicit follow-through accountability prompts (did it happen, what blocked it)
• add universal action_intent capture in pass-1/merged schemas and normalize handling so accountability works even when scenarios do not define action_intent as an outcome
• persist normalized action intent into outcomes/memory path for downstream reruns
• add tests for prompt language and action-intent capture without pass-2 outcomes

Validation

• pytest -q tests/test_reasoning_prompts.py tests/test_reasoning_execution.py

Closes #87

[DeveshParagiri]

Code Review

Verdict: ✅ Ready to merge

Summary

Implements the intent→behavior accountability loop as specified in the architecture doc §Fix 1 and Tenet 7.

Implementation

The re-reasoning prompt now includes:

## Follow Through Check

Last time I said I intended to: "{prior_action_intent}".
Did I actually do it? Be explicit:
- If yes: what exactly did I do, and what happened?
- If no: what stopped me (time, money, logistics, fear, competing priorities)?

Verification

Requirement	Status
Surfaces prior action_intent	✅
Accountability phrasing	✅
Universal capture (even without scenario outcome)	✅
First timestep handling (no prior intent)	✅
Persists to outcomes/memory	✅

Schema Changes

action_intent added as required field in both build_pass1_schema() and build_merged_schema().

No changes required.