If a pipeline declares an agent/node for the whole pipeline or for a stage, an input step will automatically hold on to an executor. As a workaround you will have to define your agent/node declaration in a way that do not wrap the input step.

I cannot think of a use case where you will ever want to hold on to an executor when waiting for a manual input. The approval process can take anywhere from minutes, days or even weeks. I think it's reasonable to say that the input step should handle the behavior internally by never run on an agent/node independent of the agent/node declaration.

Please provide use cases where holding on to an executor makes sense for better understanding in case I am missing something.