Transcript writes can be routed through a storage backend

Transcript appends, hydration, forking, snapshot copies, job timelines and history-suppression records all accept an optional storage backend and key. When one is supplied, writes go through that backend; with none supplied they fall back to a plain file append, which is what happens by default. Backend failures are re-shaped into standard errno-style errors so they show up in telemetry the same way file errors do.