[runner] Make the executor lifecycle explicit

[un-def]

GetJobInfo was not a getter. It configured runner logging and resolved the job user and working dir, and Run called the same preRun again, guarded by an "already called once" check. That guard returned nil when the first call had failed -- the log file field was already set by then -- so a second call reported success with jobUser and jobWorkingDir left unset. Only Run's jobStateHistory gate, which called this "already running or finished", kept the job from starting with those fields empty.

The pair is now Setup and Finalize, with JobInfo as a plain accessor. Setup has a single caller, so the guard goes with it. Setup finalizes itself when it fails and Run finalizes when it returns, so neither the handler nor a later caller has to remember to; the Finalize the handler defers is a safeguard and nothing more.

Finalize also takes over setting WaitLogsFinished. That used to run in a defer registered before postRun, and so ran before it. The stripper holds the last incomplete line until it is closed, which left the state that tells /api/pull its data is final published one flush early. A failed preRun was worse: it returned before postRun was deferred at all, so the buffered runner logs were never flushed and the state never became final.