v2: tea.Repaint to recover after external tty damage (ClearScreen is not enough)

[florianleibert]

Problem

tea.ClearScreen (added for #236) erases the renderer cell buffer, but it does not re-enter alternate-screen / input / cursor modes. Combined with the unchanged-view skip in cursedRenderer.flush, an app cannot recover when something outside the renderer mutates the tty.

Reproduced against v2.0.5 (and the exported API is unchanged through v2.0.9):

1. Program is in the alternate screen with a stable frame.
2. A child (or any other writer) emits ESC[?1049l. The physical terminal is now on the primary screen; the renderer still believes its full alt-screen frame is present.
3. The next content change paints only the cell delta (a few bytes). No 1049h.
4. tea.ClearScreen with a byte-identical view writes nothing on a live tick (flush returns early because !starting && viewEquals(lastView, view)).
5. The next changed frame after ClearScreen redraws text, still without re-entering the alternate screen.

Program.RestoreTerminal already proves the repair: renderer.start() replays modes from lastView, sets starting = true so the skip cannot fire, then flush emits a full frame. There is no in-loop equivalent that does this without leaving the alternate screen.

This is the same failure class #236 pointed at: a tool writes to the tty, the display is corrupt, and the app needs a forced redraw. ClearScreen never re-acquired modes, so it did not close that case.

Proposed API

func Repaint() Msg

Handled next to clearScreenMsg. The renderer, in one locked transaction:

1. reset() — drop cell-diff assumptions (fresh uv.TerminalRenderer)
2. replay modes from lastView (the body of start(), extracted so it can run under the existing lock)
3. set starting = true so the next flush cannot skip an unchanged view
4. MoveTo(0, 0) + Erase() — a fresh renderer thinks the screen is blank, so cells that are blank in our frame would otherwise leave stale content in place

A nil lastView (very early startup) is a no-op on modes plus erase.

This is not implemented by toggling View.AltScreen or writing escapes around the renderer: both leave the cell cache wrong.

ClearScreen stays as-is for wiping visual clutter in inline mode. Repaint is the recovery operation.

Tests

• ClearScreen + identical alt-screen view does not emit ESC[?1049h (the neutralized-fix baseline).
• Repaint + identical alt-screen view does re-enter the alternate screen and redraw the content.
• Proof is against renderer output bytes, not Model.View(). Send+Quit cannot prove this: flush(closing=true) already bypasses the skip.

Happy to send a PR.

[andrinoff]

Thanks for the issue! A note:

• Step 4 is already fixed in v2.0.9 by fix: pendingErase in cursedRenderer #1755. ClearScreen now sets pendingErase, so flush no longer skips an unchanged view and a full redraw happens.

As far as the issue goes, i think extending ClearScreen is the move here, rather than a new API

[andrinoff]

Can you check this out #1837?