fix: wait_for_jobs can return the wrong run because queue_agent discards the triggered run ID

[sam-saffron-jarvis]

What changed

• queue_agent now preserves the run ID returned by the jobs-v2 trigger endpoint and includes it in its JSON result alongside the job ID
• wait_for_jobs now accepts run_ids and, when provided, polls /v2/runs/{run_id} so it waits on the exact triggered run instead of whichever run is latest for the job
• kept job_ids support for backwards compatibility and added a focused test covering the exact-run polling path

Why this is high-value

This fixes a user-visible reliability bug on the queued sub-agent path. Before this change, if the same queued job was triggered again before wait_for_jobs finished polling, the waiter could attach to the newer run and return the wrong status/output. Returning and consuming the trigger's run ID removes that ambiguity and prevents result corruption.

Validation

• gofmt -w internal/tools/queue_agent.go internal/tools/queue_agent_test.go
• go build ./...
• go test ./...
• git diff --check

[SamSaffron]

Closing as requested: the high-confidence fixes have been applied or superseded in the current working tree.