fix(harness): cleanup_failed event + drop misleading exit_code capture

[HongmingWang-Rabbit]

Summary

Two self-review fixes on top of #2257 (harness hardening, just merged):

1. Drop misleading local exit_code=$? from cleanup()

trap-handler return values are ignored by bash, so capturing $? and returning it only misled a future reader into thinking exit-code preservation was happening. Removed the capture and the return $exit_code lines.

2. Surface cleanup failures as a structured event

The previous api -X DELETE ... >/dev/null 2>&1 swallowed every failure mode — including the realistic one (ADMIN_TOKEN expiring mid-run). When that happens, the workspace is leaked and there is no signal in the JSON event log.

Replaced with an -w '%{http_code}' capture:

• 200 / 204 → cleanup_deleted event with the HTTP code
• 404 → cleanup_deleted (post-condition holds — workspace absent — possibly a concurrent operator)
• anything else → cleanup_failed event with a remediation hint pointing at cleanup-rogue-workspaces.sh

Test plan

• bash -n passes
• DRY_RUN=1 exits cleanly (cleanup runs but emits cleanup_skipped because IDs are still empty)
• No behaviour change on the success path; only failure-path observability improves

Refs: #2257 review.