[rikkurea-boop]

Summary

Propagate failures from PromptedGenerator's serving backend instead of logging and silently returning.

Problem

PromptedGenerator.run() catches exceptions raised by:

llm_serving.generate_from_input(...)

but currently logs the error and returns normally.

This makes a failed generation look like a successful operator run to callers and orchestration code.

Fix

Keep the existing error logging, but re-raise the serving exception.

The production change is intentionally minimal:

except Exception as e:
    self.logger.error(f"Error during text generation: {e}")
    raise