fix: reduce recap decoy churn under tmux

[jms830]

What

Reduces the decoy-row workaround inside tmux/screen:

• keep the existing single bumpDecoy() on submit/input so vertical shifts still get a changed sentinel row;
• skip the 100ms decoy animation loop in multiplexers;
• only auto-bump the decoy during render() when card height/history length changes while inside tmux/screen.

Outside multiplexers, behavior remains unchanged.

Why

The current widget intentionally changes the hidden decoy row every render so pi-tui must repaint all rows below the recap card. That helps avoid ghost borders, but inside tmux/screen the renderer has no destructive-clear/reflow semantics for already-committed scrollback. During animation/ticking, the forced repaint of every lower row gets committed into scrollback as mid-state frames.

Observed in tmux 3.5a: repeated recap cards stacked in scrollback and progressive command rows like:

$ ls
$ ls ~/.
$ ls ~/.omp/agent/extensions/

Those artifacts are the lower rows being repainted every tick, not real shell history.

This keeps the decoy's original purpose for true vertical shifts, but stops continuous below-widget churn in multiplexers.

Testing

• Parsed index.ts, ui/status-widget.ts, and ui/status-widget.js with Bun.Transpiler.
• Applied the same patch locally via Bun patchedDependencies on an omp/tmux setup and verified it persists through bun install.
• Runtime verification target: tmux 3.5a + omp 15.12.3 + pi-recap 0.8.21/0.8.22. Expected result: no repeated recap blocks / progressive tool-row fragments in scrollback while the recap widget updates.