feat(tui): add tmux support for kitty graphics protocol

[ogulcancelik]

Adds support for rendering inline images inside tmux using the Kitty graphics protocol's unicode placeholder mode.

Problem

When running pi inside tmux + Ghostty/Kitty/WezTerm:

• Direct Kitty graphics sequences get eaten by tmux (without passthrough) or cause "stuck" images (with passthrough)
• Images orphan on screen redraws, pane switches, or scrolling

Solution

Implement the protocol's multiplexer-safe mode:

• Detect tmux via TMUX env var
• Check allow-passthrough setting before attempting image display
• Wrap sequences in tmux passthrough escapes
• Use virtual placement (U=1) with unicode placeholders (U+10EEEE)
• Encode row positions via diacritics (per official spec)

Performance Optimizations

To minimize tmux passthrough overhead:

• Transmit-once: Image data is transmitted only on first render; subsequent redraws reuse placeholders (image stays in terminal memory by ID)
• Wrap-once: All Kitty chunks are batched into a single tmux passthrough frame instead of wrapping each 4KB chunk individually
• Sync passthrough: Synchronized output markers (CSI ?2026 h/l) are passed through to the outer terminal, enabling batched rendering instead of progressive scroll

Backward Compatibility

• Outside tmux: behavior unchanged, direct placement as before
• Inside tmux without allow-passthrough: gracefully falls back to text description ([Image: foo.png 800x600])
• Inside tmux with allow-passthrough all: full image rendering

User Requirement

For images to render in tmux, users need in their tmux.conf:

set -g allow-passthrough all

[aos]

Thank you for putting this up! There's dozens of us running pi inside tmux!