Fix run_in_terminal timeout signaling and avoid foreground reuse after timeout

[meganrogge]

fix #299486

When run_in_terminal hit a model-provided timeout, it could return partial output without clearly indicating timeout, which could mislead agents and cause confusing follow-up output behavior.

Now:

1. Treats timeout as an explicit execution outcome in foreground runs.
2. Returns a clear timeout indicator in both result text and metadata.
3. Marks timed-out executions as background and detaches them from foreground session reuse.
4. Updates timeout parameter wording to state that timeout returns output with an indicator.

This allows agents to reliably detect timed-out executions, and subsequent foreground runs are less likely to be polluted by timed-out command output.

cc @Tyriar

[Copilot]

Pull request overview

This PR improves run_in_terminal timeout handling so that when a model-supplied timeout is hit, the tool explicitly signals the timeout (in text + metadata) and prevents the timed-out terminal from being reused as the foreground session terminal.

Changes:

• Updates the tool schema wording to clarify that timeouts return partial output with a timeout indicator.
• Adds an explicit “timeout” outcome to the foreground execution race and returns partial output + timeout metadata.
• On timeout, marks the terminal as background and detaches it from foreground session reuse to reduce output pollution on later runs.