perf(screen): collapse non-clear ResetPosition cursor walk-up into one CSI

[zozowell]

Problem

In the steady-state redraw path, Screen::ResetPosition(std::string&, bool) (the non-clear branch) walks the cursor back to the top-left by emitting one \x1B[1A (cursor-up) escape per row. On a tall terminal this writes a large, redundant burst of escape bytes every single frame. The cursor only needs to be repositioned once — the per-row walk encodes the same movement far more verbosely than necessary.

The change

Collapse the per-row walk-up into a single parameterized CSI cursor-up, guarded for the single-row case:

ss += '\r';  // MOVE_LEFT;
if (dimy_ > 1) {
  ss += "\x1B[" + std::to_string(dimy_ - 1) + "A";  // MOVE_UP;
}

The leading \r and the final cursor position are unchanged. On-screen output is identical.

The clear branch is intentionally left per-row: each row needs its own \x1B[2K (CLEAR_LINE) erase, so it cannot be collapsed. A code comment notes this.

Before / after (bytes per frame, non-clear path)

Rows	Before	After	Reduction
10	37	5	7.4x
24	93	6	15.5x
50	197	6	32.8x
100	397	6	66.2x

Tests

Adds src/ftxui/screen/screen_test.cpp (registered in cmake/ftxui_test.cmake):

• Non-clear output for dimy_ > 1 equals "\r\x1B[" + (dimy_-1) + "A"; equals "\r" for dimy_ == 1.
• Terminal-effect equivalence to the old per-row walk (same total rows moved up) across several heights — the byte sequences differ by design, so this compares movement, not bytes.
• The 50-row non-clear output is >= 10x smaller than the old per-row form (<= 8 bytes vs 197).
• The clear branch remains per-row.

ctest passes in full: 339/339 tests pass, including the new ones.

[ArthurSonzogni]

Thanks! That looks useful!

let's wait for the CQ.