Add multi-GPU support with --gpus flag

Adds --gpus flag to obliterate, run, and tourney commands for controlling
which GPUs to use (sets CUDA_VISIBLE_DEVICES). Works both locally and with
--remote. Models are automatically split across selected GPUs via
accelerate's device_map="auto". Also adds gpus field to remote YAML config.

examples/remote_gpu_node.yaml

@@ -9,6 +9,11 @@
 #
 # You can also use --remote on any command instead of a YAML section:
 #   obliteratus obliterate meta-llama/Llama-3.1-8B-Instruct --remote root@gpu-node --ssh-key ~/.ssh/id_rsa
+#
+# Multi-GPU: Models are automatically split across all available GPUs via
+# accelerate's device_map="auto". Use --gpus or the gpus: field to select
+# specific GPUs:
+#   obliteratus obliterate model --remote root@gpu-node --gpus 0,1,2,3
 
 model:
   name: meta-llama/Llama-3.1-8B-Instruct
@@ -39,3 +44,4 @@ remote:
   remote_dir: /tmp/obliteratus_run
   python: python3
   sync_results: true
+  # gpus: "0,1,2,3"  # uncomment to select specific GPUs (default: all)