Fix remote VM session failing with exec: trap: not found before SSH

[sbertix]

Closes #737

Summary

Opening a remote VM session over SSH failed before SSH even started, with bash: line 0: exec: trap: not found, whenever local zmx was unavailable (unbundled or over budget).

On macOS, Ghostty runs a surface .shell command string v as /usr/bin/login -flp <user> /bin/bash --noprofile --norc -c "exec -l <v>", so v must begin with an executable. ZmxAttach.buildRemoteCommand returned the bare SSHReconnectLoop.script(...) in the no-local-zmx fallback, and that script opens with the trap shell builtin. Prefixed with Ghostty's exec -l, it became exec -l trap 'exit 130' INT; ..., so exec tried to run trap as a program and the surface died with exit 127 before ssh ran. The zmx-wrapped path was unaffected because it already leads with the quoted zmx executable path.

The fallback now wraps the loop as /bin/sh -c '<loop>' (no leading exec, which would break the same way), so Ghostty produces exec -l /bin/sh -c '<loop>', leading with a real executable and running the loop in a fresh shell. This mirrors the inner /bin/sh -c the zmx-wrapped path already uses; that path is unchanged.

Type of change

• Bug fix (the linked issue is a bug report)
• Feature (the linked issue is a feature request marked ready)
• Documentation
• Other (please describe)

How was this tested?

Reproduced the failure and verified the fix at the shell level (exec -l trap ... exits 127; the /bin/sh -c-wrapped form exits 0). Added bareReconnectLoopSurvivesGhosttyExecLoginWrapping, which parses the real fallback with sh -n, runs the exact bash -c "exec -l …" shape on a benign wrapped loop, and pins the regression with a negative control on the unwrapped loop. Updated the existing fallback assertion to expect the wrapping.

• make check passes (format + lint)
• make test passes
• I built and ran the app to confirm the change works

Checklist

• This pull request is linked to an issue with Closes # above.
• For a feature, the linked issue is labeled ready.
• I am the author of this work and accountable for it; no commit is authored or co-authored by an AI agent.
• I have read the Contributing guide and the Code of Conduct.