Explain the cost of a longer context: attention is all-pairs. How does the compute scale with context length n?

prerequisite: ATTN-002