[bwachman]

Summary

• Adds a "Reject and stop" row to the permission prompt, listed separately from Reject. It declines the open call with the same non-persistent outcome as Reject, then interrupts the turn and denies any queued prompts without approving them. It never mints, widens or persists a grant.
• Shares one stop sequence between this row and Ctrl+C inside a decision gate (stopDecisionGate()), so both paths abort the reservation, close the overlay and interrupt the turn the same way.
• Keeps "Ctrl+C stop" in the permission hint line down to contentWidth 28; below that the title falls back to " Permissions".
• Reject keeps its meaning: it declines only the open call and the turn continues.

Verification

After the rebase onto main and the review fixes, CI passes on b64e31a; the results below are from the original verification.

• bun run check passes (lint, typecheck, dead exports, build, full test suite: 8576 pass, 0 fail)
• Tests pin each behavior, including "Reject and stop declines the open gate and drops queued gates", "plain Reject declines only the open gate and the turn continues", and "permission title hints keep Ctrl+C stop down to contentWidth 28"

Fixes CL-10218

[linear-code]

CL-10218

[TheGreatAxios]

Grant safety verified — Reject-and-stop is clean: the STOP outcome is always {allow:false}, shares stopDecisionGate with Ctrl+C, and queued gates are denied without approval. Four nits: the narrow-width "Ctrl+C stop" claim, an off-by-one tier comment, a weak width assertion, and an index-based test helper.

[corbits-triage]

@bwachman, please address the following:

• required checks are failing: build-and-test, test (src), eslint, prettier, typecheck, test (3/4), test (4/4), build, test (1/4), test (2/4), static-analysis