fix: fine-tune config and generator improvements

scripts/train.py: fix max_seq_len 256→320 (must match pretrained checkpoint);
increase epochs 15→50 and patience 5→10 to give the small corpus enough
gradient steps; reduce warmup 20→10 (was 22% of total steps).

scripts/generate.py: default to prepending the tonic chord when --prefix is
not given; add --no-tonic-anchor to opt out.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

scripts/train.py

@@ -54,13 +54,17 @@ TRAIN_CFG = TrainConfig(
     data_dir=DATA_DIR,
     output=CHECKPOINT,
     init_from=INIT_FROM,
-    epochs=15,
+    # Small corpus (~45 train files) → ~6 batches/epoch.
+    # 50 epochs × 6 = ~300 gradient steps; patience=10 gives a 60-step window.
+    epochs=50,
     batch_size=8,
     lr=1e-5,
-    warmup_steps=20,
+    warmup_steps=10,
+    patience=10,
     seed=42,
     device="auto",
-    max_seq_len=256,
+    # Must match pretrained checkpoint (max_seq_len=320).
+    max_seq_len=320,
 )