[AlexS-scale]

register_task_instance catches every exception and returns None, and Task.run then sets instance_id = None, even when the caller passed one. One brief store error at the start of a run means:

• the run executes with no record, though the caller (a runner, the explorer) already handed that id out as the run's link;
• retry rollbacks are refused, since there is no instance to undo through;
• ids derived from the instance id are random.

A runner wants an error here instead, so it can fail the run or let its engine retry it.

Now, with a caller-supplied instance_id, a registration failure raises out of Task.run before any step runs. Without one (the CLI, which mints none), it is still logged and the run goes unrecorded, as before. The upsert's "disappeared during re-register" error is no longer swallowed either.

Test: registering a given id when the store fails raises. It fails on main.

Confidence Score: 5/5

This PR appears safe to merge.

Summary

register_task_instance now raises registration errors when the caller supplied an instance_id.

• Runs with a supplied ID stop when instance registration fails.

_{Reviews (1) · Last reviewed commit: "fix(task): a run whose given instance id..." · Reviewed by Greptile}