fix(cli): fail sandbox exec when relay closes without exit status

[rootkiller6788]

Summary

sandbox exec seeded its exit code to 0 and only ever overwrote it when an
Exit event arrived, so an event stream that closed early, was cancelled, or
was truncated reported a successful (exit 0) run even though the command's
outcome was never observed. This makes "the command ran and returned 0" and
"we never saw what the command did" indistinguishable to callers.

Related Issue

Closes #2732

Changes

• Track whether an Exit event was observed in both exec paths
  (sandbox_exec_grpc and sandbox_exec_interactive_grpc).
• When the stream ends without an Exit event, return a diagnostic error
  instead of Ok(0), mirroring the gateway's existing exec_loop_result
  handling (Status::unavailable: "exec relay closed before the command
  reported an exit status").
• The interactive path reports the error only after the terminal is restored.

Testing

• cargo check -p openshell-cli
• Existing CLI unit tests (cargo test -p openshell-cli)

Checklist

• Change is scoped to the issue at hand
• Conventional commit + DCO sign-off
• Verified against the real code path on main

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

[johntmyers]

@rootkiller6788 please trim the title down

[johntmyers]

gator-agent

PR Review Status

The initial code review found no blocking findings: both sandbox-exec paths now distinguish a missing Exit event from an observed exit code of zero, and the interactive path restores the terminal before returning the relay error. Thanks @johntmyers — I also checked the title problem you flagged, and the PR title still includes the commit-body paragraphs instead of ending after the concise subject line.

Action required: @rootkiller6788, please trim the PR title as @johntmyers requested. The authorized E2E dispatch can proceed for this head while that metadata fix is pending.

Blocking findings:

• No blocking code findings remain.

Carried findings:

• None.

Non-blocking suggestions:

• None.

Gator metadata

• Validation: Project-valid because this focused CLI correctness fix implements accepted issue bug: sandbox exec exits 0 when the event stream ends without an exit status #2732.
• Docs: No Fern update needed because this restores the existing exit-status contract for an abnormal relay termination and adds no command, flag, configuration, or supported workflow.
• Checks: Required current-head checks have not started successfully yet; test dispatch follows this disposition.
• E2E: test:e2e required for the sandbox exec relay path and authorized by the operator.
• Head SHA: 8b645ee518028a40e6388e0936ca52e958e5f179
• Base SHA: c3993426498d5b602ec080bc1d3cbad5e8460f4a
• Merge base SHA: de4c1fecf564cd16854447ec9c08659e7b64061b
• Patch ID: 6b0668b85b7fd3b62b0059d7b3fca4806b24be5d
• Gator payload: 7
• Review mode: initial
• Previous reviewed SHA: none
• Review budget exhausted: no
• Maintainer decision required: no
• Next state: gator:in-review

[johntmyers]

/ok to test 8b645ee

[github-actions]

Label test:e2e applied for 8b645ee. Open Branch E2E Checks, find the run for commit 8b645ee, and click Re-run all jobs to execute with the label set. The run will execute the standard E2E suite after building the required gateway and supervisor images once. The matching required CI gate status on this PR will flip green automatically once the run finishes.

[johntmyers]

gator-agent

Author Follow-Up Nudge

This PR has been in gator:in-review for more than 48 business hours with the requested title cleanup still outstanding.

@rootkiller6788, please trim the PR title to the concise subject line requested by @johntmyers. If this is no longer planned, please say so and a maintainer can close it out.

[natedemoss]

Hit this same bug from the NemoClaw side — had to work around it there, so glad to see it fixed. Approach matches the gateway's exec_loop_result. I have i very similar fix if this gets closed too.

[rootkiller6788]

Thanks for the confirmation! Great to hear this aligns with what you've seen on NemoClaw.
I'll update the PR title shortly as requested.

[rootkiller6788]

Thanks for the review and for dispatching the E2E run. I've trimmed the PR title to the concise subject line (fix(cli): fail sandbox exec when the relay closes without an exit status) as requested.

On the code: both exec paths now track whether an Exit event was observed, so a stream that closes early, is cancelled, or is truncated returns a relay error instead of a spurious exit 0 — mirroring the gateway's exec_loop_result handling. The interactive path restores the terminal before surfacing that error. Branch Checks, the full E2E suite, GPU E2E, and Helm Lint are all green on the current head.

Thanks @natedemoss for the confirmation — glad to hear it aligns with what you saw on the NemoClaw side.

Could you please take another look when you get a chance?