PagedAttention：把 KV cache 分页

127        # We want to append the token to the last physical block.128        last_block = block_table[-1]129        assert last_block.device == Device.GPU130        if last_block.ref_count == 1:131            # Not shared with other sequences. Appendable.132            return None133        else:134            # The last block is shared with other sequences.135            # Copy on Write: Allocate a new block and copy the tokens.136            new_block = self.gpu_allocator.allocate()137            block_table[-1] = new_block138            self.gpu_allocator.free(last_block)139            return last_block.block_number, new_block.block_number