feat: add tqdm progbar to derisk_loopholes generation loop

No progress signal during the 6-mode quadrant (slow Qwen3-4B gen, no bar).
mininterval=maxinterval=120 per token-efficient-logging skill.

Co-Authored-By: Claudypoo <288921227+claudypoo@users.noreply.github.com>

src/projected_grpo/derisk_loopholes.py

@@ -34,6 +34,7 @@ from typing import Literal
 import torch
 import tyro
 from loguru import logger
+from tqdm import tqdm
 from tabulate import tabulate
 from transformers import AutoModelForCausalLM, AutoTokenizer, GenerationConfig
 
@@ -140,7 +141,8 @@ def run_cell(model, tok, problems, gen_cfg, mode: EnvMode, device, cell: str,
     hack_samples = []          # (problem_id, faithful_prompt_msgs, hack_completion)
     notes = []                 # exit-interview feedback (elicit cell only)
     dumped = False
-    for prob in problems:
+    # ~4-9s/problem (Qwen3-4B, group sampling). Sparse updates keep pueue logs clean.
+    for prob in tqdm(problems, desc=f"{mode}/{cell}", mininterval=120, maxinterval=120):
         prompt = _render(tok, prob["messages"], suffix)
         enc = tok(prompt, return_tensors="pt", add_special_tokens=False).to(device)
         if enc.input_ids.shape[1] + gen_cfg.max_new_tokens > 2048: