Git: repair the Vim-backed PTY commit path after ABI 43

## Why

The ABI 43 Git bottle builds, installs, and passes its noninteractive Git
operations, but the real Vim-backed `git commit` path exits 128 after the PTY
runner observes `COMMIT_EDITMSG` and sends the editor input. This is a
host/PTTY integration boundary, not evidence that Git compilation failed.

Bottle publication no longer waits on this optional integration path, so the
failure needs a separate visible platform issue rather than disappearing with
the staging workaround.

## Reproduction evidence

- Kandelo source: `5da3e9ea3b9f0f89156a21b71f08622064a1033f`
- tap source: `2134361ae63764ea593f1387e60419d555884ab6`
- hosted job:
https://github.com/kandelo-dev/homebrew-tap-core/actions/runs/31992825692/job/95279975474
- command: installed Git 2.47.1 running `git -C clone commit` through
`Kandelo/formula_support/run-pty-wasm.ts`
- result: expected 0, actual 128

The same job proves init, add, `commit -m`, clone, submodule status, shell
aliases, and the real pager before reaching this failure. The old harness did
not retain the guest transcript when the expected status differed, so the next
step is a focused local runtime reproduction or bounded per-test transcript
capture that does not rotate every Formula's staging identity.

## Completion boundary

Repair and prove the installed Git default-editor flow through Vim on both the
Node and browser host where applicable, without making this optional editor
integration a bottle-publication prerequisite again.

Start with Kandelo/formula_support/run-pty-wasm.ts and reproduce installed Git 2.47.1's `git -C clone commit` path locally, capturing the guest transcript without rotating Formula staging identity. Compare PTY behavior around `COMMIT_EDITMSG` and editor input on Node and browser hosts where applicable. Done means the Vim-backed default-editor commit exits 0 on both applicable hosts while bottle publication remains independent.

Written by the indexing model from the issue text.