Rotate Any m×n 2D Matrix

LeetCode 48 generalized to rectangles. Out-of-place rotation is trivial; in-place rotation for `m ≠ n` is the trap. Requirements Rotate a 2D matrix 90° clockwise. Square case: in-place layer-by-layer rotation; O(1) extra space. Rectangle case (NG follow-up): generally must allocate a new n × m matrix unless you allow cycle-following with bookkeeping (O(m+n) extra). Common follow-up: rotate by arbitrary multiple of 90°; reflect across an axis. Examples [[1,2,3],[4,5,6]] (2×3) rotated 90° CW → [[4,1],[5,2],[6,3]] (3×2). Notes The in-place rectangle rotation is essentially cycle-decomposition on the index permutation — explain it but don't burn time implementing. Reported follow-up: rotate inside a window of a larger matrix while keeping the rest untouched. Preparation Write square in-place rotation from memory in under 8 min. Have a 60-second explanation for why rectangle in-place is hard and what trade-offs you'd take.