[Bug]: TWG 1.2.7 emits successful output but does not exit on macOS ARM64

## Environment

- TWG CLI: 1.2.7
- Install method: direct public installer
- OS: macOS 26.6.2, Darwin 25.6.0
- Architecture: arm64
- Authentication: OAuth; `twg doctor` reports a valid token and healthy connectivity

## Reproduction

Run:

```bash
TWG_DIAG=1 twg whoami
```

The command prints the authenticated user successfully. The diagnostic output reports an HTTP 200 response from `api.atlassian.com/graphql` in under one second and a completed command profile in under one second, but the process remains alive and does not return to the shell.

The same behavior occurs with:

```bash
twg api jira:/rest/api/3/myself
twg rovo list-apps --output json
twg doctor
twg update
```

`twg update` successfully reported that it updated 1.2.5 to 1.2.7 and refreshed the skills/help cache, but the update process also remained running afterward.

## Expected behavior

After the successful response and output are emitted, the process should exit promptly.

## Actual behavior

The response is complete, but the process remains alive. Programmatic callers and agents hit their timeout and fall back to MCP or REST API access.

## Additional evidence

- Local-only commands such as `twg env auth --no-snapshot` exit normally in under one second.
- Direct `curl` requests to the Atlassian API and developer host complete in roughly 0.3–0.7 seconds.
- The following did not resolve the behavior:
- `TWG_BACKGROUND_UPDATE_CHECK=0`
- `DO_NOT_TRACK=1`
- `TWG_AGENT_DEFAULTS=1`
- `TWG_TIMEOUT_MS=1000`
- `BUN_CONFIG_HTTP_IDLE_TIMEOUT=1`
- HTTP `Connection: close`
- While stuck, the process retains established HTTPS connections to `api.atlassian.com`; process sampling shows Bun's HTTP client thread idle in `kevent`.

This appears to be a post-request shutdown or HTTP-client handle cleanup issue in the macOS ARM64 build. Please advise whether this is a known issue in 1.2.7 and whether a supported workaround or patched build is available.