Launch today's scrape for each active job and return at once.

The body is optional because the schedule does not send one: with no body this runs ACTIVE_CRON_JOBS, which is youtube only while tiktok is paused. Naming jobs in a body still overrides that, and is the only way to run a paused job such as collabstr by hand.

Each job picks up its own newest remy_cron_history payload and runs in a detached thread, so a multi-hour scrape never sits inside the request — TimeoutMiddleware would kill it at 120s and the scheduler would retry, leaving two scrapes racing. Poll the returned job_id for progress.