fix: silence num_return_sequences deprecation by baking G_s into gen_cfg

transformers warns when generation_config is passed alongside generation kwargs
like num_return_sequences. Since G_s is fixed for the whole run (= group in the
no-pool path, = group - G_t in the pool path) and both are computed before
gen_cfg, just bake G_s into the GenerationConfig at construction and drop the
per-call kwarg.

src/projected_grpo/train.py

@@ -500,7 +500,7 @@ def main(cfg: Config) -> int:
         # at T=1 dilutes them. Lower T expresses the substrate's hack propensity.
         temperature=0.7, top_p=1.0, top_k=20, min_p=0.0,
         repetition_penalty=1.0,
-        num_return_sequences=group, pad_token_id=tok.pad_token_id,
+        num_return_sequences=G_s, pad_token_id=tok.pad_token_id,
     )
 
     problems = load_problems(n_problems)
@@ -695,12 +695,10 @@ def main(cfg: Config) -> int:
                             f"live enc={live_ids[:12]}... (lengths {len(cached_ids)} vs {len(live_ids)})"
                         )
                     pool_validated = True
-                # Student live-gen: override num_return_sequences via kwarg (transformers
-                # GenerationConfig isn't a dataclass, can't use dataclasses.replace).
+                # Student live-gen. gen_cfg.num_return_sequences is baked to G_s
+                # at construction (pool path) or = group (no-pool path).
                 with torch.no_grad():
-                    out_s = model.generate(
-                        **enc, generation_config=gen_cfg, num_return_sequences=G_s
-                    ).detach()
+                    out_s = model.generate(**enc, generation_config=gen_cfg).detach()
                 # Build teacher tensor: each cached row is plen + L_t_i; right-pad
                 # to common L within the teacher batch, then F.pad to match student L.
                 teacher_seqs = [