Make `run_subtask` API support batch operation

In current implementation `SubtaskExecutionActor.run_subtask` keeps the whole lifecycle of a subtask. This disables the possibility to submit subtasks in batches, which is a major time cost of supervisors. As all calls of `SchedulingAPI.finish_subtasks` own `SubtaskResult` already, it is possible to make `run_subtask` to start subtasks only and stops tracking subtask execution.

Start by tracing SubtaskExecutionActor.run_subtask and the calls to SchedulingAPI.finish_subtasks to understand the current lifecycle and result ownership. Done means run_subtask can start subtasks without tracking their execution, while batch submission and finish_subtasks continue to handle SubtaskResult values correctly.

Written by the indexing model from the issue text.