[ERR_HTTP2_INVALID_SESSION]: The session has been destroyed — causes repeated transient retries

### Describe the bug

The CLI surfaces Request failed due to a transient API error. Retrying... followed by:

✗ Error [ERR_HTTP2_INVALID_SESSION]: The session has been destroyed

This happens frequently (multiple times per session, often mid-turn on long reasoning responses). I would then be unable to issue a /restart command, and have to quit the cli and relaunch is with resume to continue. And then would encounter the issue again frequently

### Affected version

1.0.47

### Steps to reproduce the behavior

Steps to reproduce

1. Launch copilot in a normal repo.
2. Run any non-trivial prompt that produces a long reasoning turn (e.g. multi-file edit, /research,
planning).
3. Within the turn, the HTTP/2 session error fires; CLI retries.

The error is non-deterministic but reproduces within ~5–15 minutes of active use.

What I've already tried

- /restart — temporarily clears, error returns later.
- /update — already on latest (1.0.47).
- Switched models via /model — happens on multiple models.
- Forced IPv4 to rule out a broken IPv6 path: NODE_OPTIONS="--dns-result-order=ipv4first" copilot

Verified the flag is applied. Error still occurs.

### Expected behavior

_No response_

### Additional context

OS macOS 15.7.7
arm64 (Apple Silicon)
Node version v22.22.0
terminal: cmux
shell: zsh