model_hub: populate_placeholders silently passes unresolved {{tokens}} to LLM on column resolution failure

Summary

populate_placeholders() catches all non-media exceptions and returns the original (unsubstituted) messages as a fallback. Unresolved {{column_name}} tokens reach the LLM verbatim with no error surfaced to the caller.

File: futureagi/model_hub/views/run_prompt.py:460-467

    except Exception as e:
        if media_error:
            raise e
        else:
            traceback.print_exc()
            logger.exception(f"Fatal error processing messages: {e}")
            # Return original messages as fallback
            return messages

The inner per-column loop (lines 403-407) also continues silently on column resolution errors.

Impact

• LLM receives {{column_name}} as literal text when a column is missing, a cell is null, or column lookup fails.
• The prompt response is subtly wrong with no error surfaced to the user.
• Callers cannot distinguish "all resolved" from "fallback due to exception" — both return a list of messages.
• Parallel to simulate issue #312.

Suggested fix

After template rendering, scan for remaining {{...}} tokens in the output messages. If any are found, either:

• Return an error status (preferred), or
• Log at WARNING with the list of unresolved tokens.

See also

ADR 024 in docs/adr/024-populate-placeholders-no-unresolved-validation.md.
Simulate issue #312 for the parallel problem in the simulate module.

Start in futureagi/model_hub/views/run_prompt.py at populate_placeholders(), especially the per-column loop around lines 403-407 and fallback around lines 460-467. Read docs/adr/024-populate-placeholders-no-unresolved-validation.md and compare the parallel simulate issue #312. Done means unresolved {{...}} tokens cannot reach the LLM silently and the caller receives or can identify the failure.