[eersnington]

Refs rivet-dev/rivet#5958.

A ctx.race with a ctx.sleep branch never timed out in-process work: the race waited for every branch to settle, so a slow branch won after the deadline had passed.

• executeRace keeps a yielded branch and runs it again at its deadline while other branches still run.
• The race writes its earliest deadline with driver.setAlarm, so an evicted run wakes again at that deadline.
• When a branch wins, the race stops waiting for cancelled steps. Their late results are discarded and are not recorded as failures or retried.
• Eviction does not interrupt a running step, so the step's result is still recorded.
• Second commit: the race saves the winner before it deletes the losing branches, through one queued save. Before, a crash between the delete and the save lost the winner, and replay ran confirmed steps again.

	main	This PR
2000 ms step raced against ctx.sleep(300)	step wins at ~2000 ms	timeout wins at ~300 ms
Crash between deleting losers and saving the winner	confirmed loser step runs again	winner kept, no re-run

Same fix in engine-ee: rivet-dev/engine-ee#1777.

---

This is part 4 of 5 in a stack:

•  5  fix(workflows): stop long step timeouts from firing immediately #28
•  4  fix(workflows): wake race branches at their deadlines while other branches run #27 👈
•  3  fix(workflows): reject removed try scopes #26
•  2  fix(workflows): mark settled history as visited on replay #25
•  1  fix(workflows): save overlapping join history in order #24