[runner] Mark logs finished only once the executor has finished

[un-def]

Previously, Server.stop() set WaitLogsFinished as soon as /api/stop arrived. /api/pull treats the first request it serves in that state as the final one and closes pullDoneCh -- the barrier that keeps the runner alive until the server has fetched the job's last logs and state.

The job is still stopping at that point. The command has up to killDelay before it is killed, so the state being served is not final, and the barrier is satisfied before the state it is meant to protect exists:

08:38:39  /api/stop
08:38:41  /api/pull                     <- closes pullDoneCh here
08:38:51  Job state changed new=terminated
08:38:51  Job finished, shutting down
08:38:51  Logs streaming finished endpoint=/api/pull   <- 92us later

The runner exited 64ms after computing the terminal state, without ever serving a pull that carried it. The server's next poll found nothing listening, reported the instance unreachable, and terminated the job on the disconnect timeout two minutes later -- discarding the state the runner had worked out.

Now stop() only cancels the run. The executor sets WaitLogsFinished from its own defer when Run returns, after the terminal job state has been recorded, so the first pull to observe it carries that state and the runner exits only once the server has taken it.

HasMore is derived from the same state and was likewise reported as false for the whole stop window. Nothing observed it: the server's PullResponse has no such field.