[Coordinator] API `get_task` should re-deliver the assigned task instead of return error

[noel2004]

Considering a prover with public key P, after coordiantor has assigned a task to the prover and has not completed it yet, another call to get_task api would return an error.

Though the proving-sdk is supposed to persist the assigned task in local. There would be many cases that it missed the record (for example, the db data is removed by accident, or program crash before task is persisted in some testing running) and cause an error can not be recover in short time (unless we have changed the pk of prover, or wait until current task has expired).

Also, we notice that if coordinator simply return the task it has assigned, persisting of task in prover side is not needed, reducing complexity in development and deployment.