fix: clone grammar bias per step in generate_period

_grammar_bias returned a shared module-level singleton that the loop
mutated in place (EOS block + repetition penalty). The penalty thus
accumulated across positions within a call and persisted across calls,
collapsing output to HOLD/NC until process restart. Clone the bias each
step so edits stay local. Add regression tests guarding the invariant.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

src/generate.py

@@ -322,7 +322,10 @@ def generate_period(
                 break
             inp = torch.tensor([ids], dtype=torch.long, device=device)
             logits = model(inp)[0, -1]                        # [vocab_size]
-            bias = _grammar_bias(last_id, pos_in_bar, positions_per_bar)
+            # _grammar_bias returns a shared module-level singleton; clone before
+            # any in-place edit so EOS-blocking and the repetition penalty never
+            # leak across positions or across calls.
+            bias = _grammar_bias(last_id, pos_in_bar, positions_per_bar).clone()
             if n_bars is not None and bars_completed < n_bars:
                 bias[_EOS] = float("-inf")   # don't let the model stop early