Copilot Chat can act on truncated tool-result previews instead of the full payload

## Problem

When Copilot Chat receives a large tool result that is truncated in the chat UI and saved to a linked file, it can appear to reason from the displayed preview instead of opening the full payload. That can lead to incorrect conclusions on large results such as project-cache responses, logs, and long JSON payloads.

## Steps to reproduce

1. Trigger a tool call that returns a large result in Copilot Chat.
2. Let the response be truncated so the UI shows only a preview and a linked file for the full content.
3. Ask Copilot Chat to summarize or act on the result.
4. Observe that it may use the visible excerpt rather than the full payload.

## Expected behavior

Copilot Chat should use the full tool payload, or explicitly open/read the linked file, before making assertions.

## Actual behavior

Copilot Chat can continue reasoning from the truncated inline preview as if it were complete.

## Impact

This can produce incorrect reasoning on large tool outputs and force unnecessary fallback to browser verification or other sources.

## Notes

I observed this while working with a large project-cache issue payload in VS Code Chat. The payload was present, but the chat UI showed only a preview and I had to follow the linked file to see the full content.