feat: add benchmark orchestration, LIBERO-plus install parity, and eval hardening

- Add lerobot-benchmark CLI for multi-benchmark train/eval workflows
- Add benchmark_training.mdx documentation
- Add libero-plus pip extra alias with EGL probe deps matching standard libero
- Harden libero.py: wand mock, init-state fallback, renderer EGL→OSMesa fallback
- Add multimodal_analysis.py script and SLURM training template

Made-with: Cursor

scripts/train_smolvla_libero_plus.slurm

@@ -0,0 +1,29 @@
+#!/bin/bash
+#SBATCH --job-name=smolvla_libero_plus
+#SBATCH --partition=hopper-prod
+#SBATCH --nodes=1
+#SBATCH --ntasks-per-node=1
+#SBATCH --gpus-per-node=4
+#SBATCH --cpus-per-task=48
+#SBATCH --mem=200G
+#SBATCH --time=12:00:00
+#SBATCH --output=logs/smolvla_libero_plus_%j.out
+#SBATCH --error=logs/smolvla_libero_plus_%j.err
+
+set -euo pipefail
+
+eval "$(conda shell.bash hook 2>/dev/null)"
+conda activate lerobot312
+
+cd /admin/home/pepijn/lerobot_wt_robocasa
+
+lerobot-benchmark train \
+    --benchmarks libero_plus \
+    --policy-path lerobot/smolvla_base \
+    --hub-user pepijn223 \
+    --num-gpus 4 \
+    --steps 30000 \
+    --batch-size 32 \
+    --eval-freq 0 \
+    --wandb \
+    --dataset.repo_id=pepijn223/libero_plus_lerobot