Orbit: add a toggle to suppress agent-generated job-monitoring scripts

A tester reports that most of the time they don't want Orbit to monitor Galaxy for job completion, and would like a toggle (or similar) to tell the agent not to write monitoring scripts.

**Observed behavior**
During a batch tool run (launching a set of alignment jobs over collections), the agent repeatedly wrote Python polling scripts -- a `monitor_*.py`-style sweep -- to watch job and collection states until everything finished. That babysitting is the behavior the user wants to be able to turn off.

**Why it matters**
For many workflows the user would rather fire-and-check than have the agent spin up polling loops. The monitoring scripts add noise (and token/cost), and aren't wanted by default.

**Proposed**
A user-facing toggle / setting (or an easily discoverable per-prompt directive) that tells the agent not to create job-monitoring/polling scripts. When off, the agent should launch the work and report what it kicked off, leaving monitoring to the user or to an on-demand status check.

**Related**
#215 tackles the agent *blocking* on `sleep` loops by proposing an async invocation-wait primitive + idle re-nudge. This is the complementary control angle -- an off-switch for monitoring rather than a smarter wait. The two should probably be designed together.

**Environment:** Orbit 0.2.0, Gemini 3.5 Flash.

No implementation files or tests are named. Start by locating the Orbit path that generates monitor_*.py polling scripts, then read the related async-wait discussion in #215. Done means a discoverable setting or directive prevents monitoring-script creation, launches the work, and reports what was started.

Written by the indexing model from the issue text.