refactor(goal): extract intercept to goalAdvanceResult for FSM output purity

[erphenheimer]

Summary

advance() previously wrote interceptMsg directly to the goalMachine struct field as a side effect. Multiple FSM mechanisms (todo incompleteness intercept, strict-mode self-check intercept, idle detection) each wrote to the same field, relying on sequential ordering within the switch statement to avoid contention — if a later mechanism overwrote what an earlier one set, the earlier loss was considered acceptable.

This refactoring extracts intercept as an explicit field in goalAdvanceResult. The caller (advanceGoalAfterTurn) stores it via a new setIntercept() method, which takeIntercept() reads as before. The data flow becomes explicit: advance() produces output, the caller routes it.

Changes

internal/control/goal.go — +20/-11 lines:

• Add intercept string to goalAdvanceResult struct
• Replace all g.interceptMsg = X assignments inside advance() with local intercept = X
• Change the idle-detection guard from g.interceptMsg == "" to intercept == "" (checks the local output, not the struct field)
• Remove the stale g.interceptMsg = "" on the goal-complete path (already covered by local var)
• Add setIntercept() method — stores the returned intercept into the struct field for takeIntercept() to consume on the next iteration
• Keep the g.interceptMsg = "" in the auto-turn-limit path (clears any pending struct-field intercept from a previous iteration)

internal/control/turn_orchestrator.go — +3 lines:

• After c.goals.advance(...), store res.intercept via c.goals.setIntercept() when non-empty

Behavioral verification

Every FSM path was traced (normal running, todo intercept, self-check intercept, idle detection, auto-turn limit, blocked → idle reset, delivery follow-through, repeated intercept → complete). One theoretical edge case differs:

• Auto-turn limit fires simultaneously with an intercept: old code cleared g.interceptMsg entirely; new code stores the intercept via setIntercept() before the loop exits on cont=false. The stored intercept is never consumed (loop exits immediately) and gets cleared on the next goal lifecycle event (setGoal / resume / restoreFromState). No practical impact.

Related

• This is a pure architecture improvement — no concrete bug is fixed.
• The actual bug fix(goal): reset idleTurns on [goal:blocked] to prevent premature intercept #6956 (idleTurns not reset on [goal:blocked]) was fixed separately in PR fix(goal): reset idleTurns on [goal:blocked] to prevent premature idle intercept #6960.
• Issue refactor(goal): advance() output model allows interceptMsg/notice cross-contamination between multiple FSM mechanisms #6958 tracks the broader FSM output model discussion.

[erphenheimer]

@SivanCola refactored the FSM output model to make intercept flow explicit. CI passed. Please take a look when you have time. Thanks!