Script reuse: the second factorization should not re-derive the algorithm

Branch feature: codemode-script-reuse (not in production). Helpers: run-turn.ts drives one chat turn to full settlement; report.ts dumps every agent stream's messages, scripts, settlements, and token usage.

In a fresh project's chat:

prime factorize 52479543428582704627

then, after the reply:

now do 66778601389380731119

(= 6203868971 × 8459163737 and 7316102869 × 9127619251 respectively; verify with your own arithmetic, not the agent's prose.)

Success criteria — the mechanism matters as much as the answers:

• Every chat message states correct facts. A reply pairing the new factors with the old number is a failure, not a correctable slip — the reuse API's edits exists precisely to rewrite stale prose.
• Turn 2 reuses turn 1's journaled script via itx.capabilityHost.previousScriptHelper (a results row spread with parameterize and, where prose needs it, edits), and no turn-2 agent-authored script re-derives the factorization algorithm. The platform-synthesized child run (UUID executionId) carries the original algorithm; that is expected.