[agent] The nightly e2e-extended legs e2e_bun_lockb (bun 1.0.36 and 1.1.45) have been red on main since #1648 merged.

Evidence

• Run https://github.com/SocketDev/socket-patch/actions/runs/40639505834 (schedule, main b787454, 2026-10-28), jobs 122654524514 and 122654524565.
• native_binary_hosted_vendored_takeover_roundtrip panics at crates/socket-patch-cli/tests/e2e_bun_lockb.rs:928: expected redirect_revert_failed, but vendor --offline --dry-run now returns the top-level error offline_eject_unavailable ("ejecting 1 hosted package needs 1 patch record(s) from the Socket API, and this run is offline; re-run without --offline").
• Reproduces locally with bun 1.0.36 on origin/main.

Cause
#1648 (vendor hosted patches directly from lockfiles) made vendor --offline refuse up front whenever a hosted pin is in scope, which its description records as the chosen behaviour. This test still expects the older per-pin redirect_revert_failed event. It only runs on the bun 1.0/1.1 legs, which are nightly-only, so the PR and merge-queue runs never exercised it.

Fix
Update the test's offline-refusal step to the new contract (exit 1, offline_eject_unavailable, nothing written), and check that the rest of the roundtrip still holds.

Generated by Claude Code