Missing load option in runner

Right now saving works on the agent level, but to continue training is should probably also be enabled on the runner level. Likely as simple as calling agent.load() if a loading flag is set.

Start by locating the runner implementation and the existing agent-level saving and loading behavior. Trace how runner options are defined and how training resumes, then verify that enabling the loading flag causes the agent to load before continuing training.

Written by the indexing model from the issue text.