Agent module async load support

[vkruoso]

Support has been added to load async agents on the Eval framework.
commit

I would like to understand why it would be a problem to add this to the core agent loading mechanism of the framework, more precisely in the google/adk/cli/utils/agent_loader.py file.

I understand this might be a design decision. My use case is that one agent definition (along with its tools and so on) is loaded by an internal HTTP call. This allow us to dynamically load the agent on each "run" based on the run parameters. If that is not the best solution for dynamic agents, how would one approach to do it with ADK?

I see an example for MCP tools with get_agent_async being used, but that implies that we do not use the standard fast api server.

Any hints on that?