Refactor undo cursor

[zxch3n]

This change internalizes some logic related to cursor transformation when performing undo/redo. It makes the API much simpler.

Specifically, the recorded cursor info for each undo/redo item is different when we're inside an undo/redo loop.

// When in the undo/redo loop, the new undo/redo stack item should restore the selection
// to the state it was in before the item that was popped two steps ago from the stack.
//
//                          ┌────────────┐
//                          │Selection 1 │
//                          └─────┬──────┘
//                                │   Some
//                                ▼   ops
//                          ┌────────────┐
//                          │Selection 2 │
//                          └─────┬──────┘
//                                │   Some
//                                ▼   ops
//                          ┌────────────┐
//                          │Selection 3 │◁ ─ ─ ─ ─ ─ ─ ─  Restore  ─ ─ ─
//                          └─────┬──────┘                               │
//                                │
//                                │                                      │
//                                │                              ┌ ─ ─ ─ ─ ─ ─ ─
//           Enter the            │   Undo ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─▶   Push Redo   │
//           undo/redo ─ ─ ─ ▶    ▼                              └ ─ ─ ─ ─ ─ ─ ─
//             loop         ┌────────────┐                               │
//                          │Selection 2 │◁─ ─ ─  Restore  ─
//                          └─────┬──────┘                  │            │
//                                │
//                                │                         │            │
//                                │                 ┌ ─ ─ ─ ─ ─ ─ ─
//                                │   Undo ─ ─ ─ ─ ▶   Push Redo   │     │
//                                ▼                 └ ─ ─ ─ ─ ─ ─ ─
//                          ┌────────────┐                  │            │
//                          │Selection 1 │
//                          └─────┬──────┘                  │            │
//                                │   Redo ◀ ─ ─ ─ ─ ─ ─ ─ ─
//                                ▼                                      │
//                          ┌────────────┐
//         ┌   Restore   ─ ▷│Selection 2 │                               │
//                          └─────┬──────┘
//         │                      │                                      │
// ┌ ─ ─ ─ ─ ─ ─ ─                │
//    Push Undo   │◀─ ─ ─ ─ ─ ─ ─ │   Redo ◀ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ┘
// └ ─ ─ ─ ─ ─ ─ ─                ▼
//         │                ┌────────────┐
//                          │Selection 3 │
//         │                └─────┬──────┘
//          ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ▶ │   Undo
//                                ▼
//                          ┌────────────┐
//                          │Selection 2 │
//                          └────────────┘
//
// Because users may change the selections during the undo/redo loop, it's
// more stable to keep the selection stored in the last stack item
// rather than using the current selection directly.

[Leeeon233]

It looks much simpler