Stale-checkout pre-flight block has no self-heal: runs fail hard instead of pulling main themselves #802

Related to #801.

When a coding run's local checkout is behind origin/main, the pre-flight check in #801 correctly refuses to start the run so it doesn't build on a stale base. But the block is a complete dead end: the run fails immediately with stopReason 'failed' and a message telling the human to run git pull --ff-only directly on the machine hosting the agent.

This happened twice in a row today on the ProAgentStore/platform coder itself (Sergeys-Mac-mini runner), 19 commits behind origin/main (local 5b671b2, origin/main 0af538a). Passing an explicit 'pull first, then do the work' instruction in the objective does not help — the pre-flight check stops the run before the objective is even attempted, so the agent never gets a chance to self-heal by pulling.

Requested fix: when the only problem is a stale-but-clean checkout (no uncommitted local changes, no conflicts), let the pre-flight step attempt git pull --ff-only itself before blocking, and only fail hard if the pull itself fails (e.g. diverged history, conflicts, dirty working tree). This removes the need for a human to physically intervene on the runner machine just to unblock a routine stale checkout.