Copilot chat terminal command completed notification disrupts chat flow/context with "What would you like me to work on"

```
Version: 1.137.0 (user setup)
Commit: 645f29cc3176500b4b5762ba887cf2a7f0ffdf2c
Date: 2026-09-08T13:43:59-07:00
Electron: 42.10.0
ElectronBuildId: 15109253
Chromium: 148.0.7778.280
Node.js: 24.18.1
V8: 14.8.178.38-electron.0
@github/copilot: 1.0.84.70.gdb75d0d
@github/copilot-sdk: 1.0.13-preview.4
OS: Windows_NT x64 10.0.26200
```

Running in local Agent mode
Model: GPT-5.6 Luna High 1M context via Github Copilot

## Description

When an agent launches a run_in_terminal mode=async and interacts with it via send_in_terminal, sometimes the terminal command completed notification will come in and interrupt the current turn and print a "What would you like me to work on" message devoid of the prior conversation context.

In the above example, the last turn send_in_terminal sent a Ctrl+C signal and explictly told the agent not to end the turn:
```
Note: The input you sent was a cancel signal (Ctrl-C / Ctrl-D / Ctrl-\). The previously running command was interrupted, not completed. This is not a signal to end the turn — if you intend to run a recovery or follow-up command, issue it now in this same turn. Call get_terminal_output first if you need to verify the shell is back at a prompt.
```

And yet, the turn appears interrupted by the next "User Message":
```
[Terminal 37076e3a-0af5-4dcd-9536-06c32a5bfb28 notification: command completed. Use send_to_terminal to send another command or kill_terminal to stop it.]
```

The model request seems to lack prior context (note the substantial decrease in input token count), and the agent then responds with a response like "What would you like me to work on in the xxx workspace?"

Follow up conversations also seem to continue to have the reduced input token counts making it seem that the prior context is permanently lost/compacted. The UI also has no ability to restore conversation/fork to before the terminal notification to get that context back.