Pruning LLMs Like a Physicist: Block Removal as an Ising Optimization Problem