fix(pet): support iTerm2 runtime and vertical alignment

[thegreatesthoneybee]

Summary

• reapply Gajae Pet iTerm2 rendering on top of v0.13.3
• keep iTerm2 Pet visible when GJC is launched through the local launcher or SSH
• move the iTerm2 Pet upward by 0.25 terminal cell using transparent image padding
• include the iTerm2 launcher and runtime cleanup behavior

Verification

• bun test packages/tui/test/gajae-pet.test.ts packages/coding-agent/test/gajae-pet-widget.test.ts
• 47 tests passed
• bun --cwd=packages/coding-agent run check:types
• runtime smoke test with gjc/0.13.3, /pet red, iTerm2 payload, and height=50px canvas verified

Note: the full coding-agent type check still reports pre-existing SDK reconnect option type errors outside the Pet changes.

[Yeachan-Heo]

Disposition: superseded by the current-dev iTerm Pet repair

Thanks for carrying the iTerm2 runtime/alignment follow-up. This PR cannot be integrated as submitted:

• It targets main at the frozen 0.13.3 snapshot (accd043c) rather than the integration branch. Its history is more than 1,100 dev commits behind, so retargeting would expose the entire release-line divergence rather than a bounded four-file patch.
• Its three product files (gajae-pet-widget.ts, gajae-pet-widget.test.ts, and the coding-agent changelog) overlap the already-active current-dev repair PR fix(pet): stabilize iTerm2 Gajae Pet placement, drag handling, and animation rendering #4591, which owns the same iTerm placement/runtime contract and now includes geometry-derived placement, drag-path handling, non-destructive capability probing, multiplexer protocol guards, raster-lease animation ownership, stopped-TUI suppression, timeout ordering, and focused lifecycle tests.
• The added run-gjc-pet.sh forces GJC_FORCE_IMAGE_PROTOCOL=iterm2 by default through local/SSH or multiplexer launches. That conflicts with the current safety contract restored in fix(pet): stabilize iTerm2 Gajae Pet placement, drag handling, and animation rendering #4591: raw image protocols must not be forced through tmux/screen/zellij except the explicitly supported path.

The useful problem signal is being retained in #4591 on current dev; maintaining two overlapping implementations would make lifecycle regression attribution ambiguous. Closing this stale release-snapshot variant as superseded rather than asking for another rebase/reimplementation cycle.

REQUEST_CHANGES / terminal disposition: close as superseded by #4591; no commits from this branch should be merged independently.

—
[repo owner's gaebal-gajae (clawdbot) 🦞]