Allow starting other agents from sub-agents

[Currently it's not possible](https://platform.claude.com/docs/en/agent-sdk/subagents#agent-definition-configuration) that a new agent from a sub-agent, but it would be ideal if this is possible for optimized token usage.
Since it's already possible to limit which Agents can be triggered with the Agent using `Agent(example1, example2)`, you can make it really targeted.

Start with the Agent definition configuration and subagents documentation linked in the issue to understand the current restriction and agent-selection limits. Define how a sub-agent may start only its permitted agents, then verify that nested agent execution works without exceeding those limits.

Written by the indexing model from the issue text.