fix(chat): thread the abort signal into local turns so Ctrl+C works [SC-A1.1]

[AetherAI3]

Problem

Ctrl+C did nothing to a local turn.

The REPL builds an AbortController per turn and aborts it on Ctrl+C. runTurn accepted that signal — and then dropped it:

const backend = await resolveBackend(ctx);
if (backend === "local") {
  await runLocalTurn(ctx, prompt);   // ← signal not passed
  return;
}
await runCloudTurn(ctx, prompt, signal, onFrame, onPulsePaint);   // cloud gets it

The abort fired and nothing was listening. The turn ran to completion regardless. Only the cloud path was ever cancellable.

Contract

runLocalTurn takes the signal and closes the brain on abort. close() is what unblocks a loop parked on a tool result, so an abort arriving mid-turn is observed rather than waiting the turn out.

The listener is registered before the loop starts, so a signal that is already aborted is honoured rather than starting work that was cancelled before it began.

An aborted turn returns rather than throwing. The user asked for the stop; it is not a failed turn and must not be reported as one.

Implementation

Adds a LocalTurnDeps seam ({ brain, exec }) mirroring the one smoke.ts already uses, so the abort path is testable without an Ollama server, a child process, or real tool execution. runLocalTurn is exported for the same reason.

Tests

4 added, driving a brain that emits one tool_call and then parks — exactly as the real one does while awaiting sendToolResult.

test	proves
aborting closes the brain	the signal reaches something that can act on it
an already-aborted signal stops the turn	no work starts that was cancelled first
an aborted turn does not reject	a user-requested stop is not an error
no signal still completes normally	the happy path is unchanged

Mutation-checked — and the failure mode is the point. Restoring the old behaviour does not fail the tests, it hangs them:

	result
signal dropped (old behaviour)	killed by timeout, exit 124, zero TAP summary lines — the turn never settles
signal wired (this PR)	exit 0, 4/4 pass

That hang is precisely what a user experienced when they pressed Ctrl+C.

Gates at 8ca821d:

command	result
npm run typecheck	exit 0
npm test	926 pass / 0 fail

Baseline on clean 41a7e261, measured in the same session: 922 / 0.

Scope — what this does NOT fix

This makes a local turn cancellable between steps: during a model request, or while parked awaiting a tool result.

It does not interrupt a tool that is already executing. tool_executor.ts:88 still uses blocking spawnSync with no AbortSignal, so:

• Ctrl+C during a long run_tests is still not observed until that command returns.
• A timed-out command still orphans its children — spawnSync signals only the direct child (cmd.exe / sh), so the actual npm test / pytest keeps running. Near-certain on Windows.
• The whole event loop is blocked for the duration, so heartbeats and animations freeze too.

Why that is a separate slice, not grafted on here: fixing it needs spawn() with process-group cleanup → ToolExecutor.run must become async → execute() can no longer serve run_shell/run_tests synchronously → finalVerify must become async. That is the synchronous ground-truth gate (verify_gate.ts:71, called from code.ts:371, with 13 tests pinning its behaviour, including the "a brain can never upgrade a red run" contract).

Turning the verification gate async deserves its own reviewable change rather than riding along behind a three-line signal fix.

Security notes

• No change to what is executed, or to permission evaluation — this only adds a cancellation path.
• No new process spawning, no new filesystem or network access.
• The abort listener is removed in finally, so a long REPL session cannot accumulate listeners on reused signals.

Known limits

• The blocking-spawnSync items above.
• chat.ts still re-implements the host loop rather than sharing code.ts's, and the two have already drifted: different permission-prompt fields (chat.ts reads url/query, code.ts does not), different truncation widths, different denial strings, and no finalVerify on the REPL path at all. Loop convergence is a later slice of this lane.
• hostLoop still has no turn or wall-clock budget for LocalBrain/CloudBrain.

Dependency and merge order

Independent of #72, #73, #74, #75 — branched from origin/main.

Touches src/commands/chat.ts, which SC-A4 (#75) deliberately avoided for this reason. #75 touches only brain_ollama.ts, so there is no overlap; either order works.