improvement: drop jobId from hosted training run response

[JannikSt]

• Removes job_id / jobId from HostedTrainingRunResponse and the prime rl dispatch surface (JSON output and human print).
• job_id is a platform-internal scheduling id; the backend is dropping it from POST /api/v1/training/runs so clients can't infer the cluster from it.
• Beta-gated, no compat shims needed. run_id and token_value remain.

---

Note

Low Risk
Narrow CLI/API contract change for a beta-gated endpoint; automation that parsed jobId from dispatch JSON would break, but runId remains.

Overview
Aligns the Prime CLI with the backend removing jobId from POST /v1/training/runs, so clients no longer see the platform-internal scheduling id (which could leak cluster inference).

HostedTrainingRunResponse no longer parses job_id; run_id and token_value are unchanged (and token_value is still withheld from stdout/JSON on dispatch).

Full-FT dispatch via _dispatch_full_finetune_run now returns JSON {"run": {"runId": ...}} only and prints a single Dispatched line with run_id, without the job id suffix.

^{Reviewed by Cursor Bugbot for commit 1ff0c1f. Bugbot is set up for automated code reviews on this repo. Configure here.}