fix: print tracebacks for per-item failures in batch generation loops

[smcshahid]

What

Adds traceback.print_exc() before each of the three per-item except Exception: continue handlers in infer_helios.py's batch modes (prompt_txt / image_prompt_csv / interactive_csv), plus the import traceback.

Why

A failing item currently leaves no trace: the process exits 0 with fewer (or zero) outputs after minutes of GPU time. Observed in practice: an interactive run OOM'd at the final postprocess after ~10 minutes and was indistinguishable from success at the process level — echo $? → 0, output folder empty.

Compatibility / regression

• Keeps the author-intended skip-and-continue batch semantics; smallest possible change.
• stderr output only; no API/behavior/exit-code change.
• Deliberately separate from the interactive chunk-count fix (unrelated concerns); together they turn that failure from "exit 0, no files" into a visible OOM with a traceback.

🤖 Generated with Claude Code