Phase 6.4.25b: version the corrective LoRA #2 adapter (out/lora-adapter-penpot-v2/)

171 steps, train_loss=0.6808, eval_loss=0.6348 (comparable to v1's 0.6306),
peak CUDA memory 74.58GB -- identical to v1's run, confirming the 13
corrective seeds didn't change the memory/length envelope. All guards
verified in adapter_config.json: use_rslora=false, use_dora=false,
lora_bias=false, modules_to_save=null, r=32, lora_alpha=64.

v1 adapter (out/lora-adapter-penpot/, commit 363648f) stays untouched as
a rollback point. New OUTPUT_DIR avoided the overwrite guard.

adapter_model.safetensors sha256 verified identical before/after copying
out of the container (root-owned file, fixed ownership via chown inside
the container since this OUTPUT_DIR landed inside the worktree, unlike
v1's which was copied in from outside).

.gitignore

@@ -13,6 +13,8 @@ out/lora-adapter/checkpoint-*/
 !out/lora-adapter/training_args.bin
 !out/lora-adapter-penpot/adapter_model.safetensors
 !out/lora-adapter-penpot/training_args.bin
+!out/lora-adapter-penpot-v2/adapter_model.safetensors
+!out/lora-adapter-penpot-v2/training_args.bin
 __pycache__/
 *.pyc
 .ipynb_checkpoints/
@@ -22,7 +24,9 @@ __pycache__/
 # Fase 6: checkpoints intermedios del LoRA #2. El adapter final
 # (out/lora-adapter-penpot/adapter_model.safetensors, ~169 MB) SI se commitea, igual que el de
 # la Fase 3; los checkpoint-*/ del Trainer son decenas de GB y viven solo en spark.
+# out/lora-adapter-penpot-v2/ es la iteracion correctiva del paso 6.4.25b (mismo trato).
 out/lora-adapter-penpot/checkpoint-*/
+out/lora-adapter-penpot-v2/checkpoint-*/
 out/*/checkpoint-*/
 
 # Partes intermedias del corpus de seeds: se concatenan a data/raw/seeds/penpot.jsonl, que es