[ problem ] Stack traces within the system are hard to read

[LukeLalor]

This is due to passing generators around for streaming events (ie, context within apu generated by tool calls). Likely this fix will primarily revolve around aggregating stacks when handling errors that bubble up from these generators.

Repro

execute llm request with bad model Stack trace of reported error in server logs does not make sense

Success Criteria

intentional stack trance produced in the following scenarios...

• [x] Bad yaml file server startup
• [x] bad agent configuration server startup
• [x] server error creating process
• [x] internal agent error on action
• [x] error within apu / llm unit
• [x] error making request to llm
• [x] error when retrieving / parsing