fix: prevent container-name collision for id-less concurrent jobs

execute_job derived a missing job id from a bare millisecond timestamp,
and that id becomes the Docker container name. Two id-less jobs racing
within the same millisecond produced identical `tako-<ts>` names, so the
second `docker run` failed with a name conflict and the job was lost.

Append a short random suffix to the id-less fallback. Explicit ids are
still honoured verbatim.