Vulnerability: HITL Bypass via UI Truncation (Vertical Newline Injection) in Command Confirmations

### Summary

There is a critical Human-in-the-Loop (HITL) bypass vulnerability in the CLI's command execution confirmation prompt. An attacker (or a compromised/malicious agent response) can use **Vertical Newline Injection** to execute UI redressing.

Currently, if an agent proposes a massive shell command that is heavily padded with newlines (e.g., hiding a payload like `curl | bash` at the very bottom), the terminal correctly truncates the view using `MaxSizedBox`. However, the `ToolConfirmationMessage` logic remains completely disconnected from this truncation state.

As a result, the "Allow" confirmation options remain active, allowing a user to accidentally approve a command where the actual malicious payload is hidden from their terminal view.

### Impact

Users relying on the HITL confirmation step might unknowingly authorize hidden malicious shell commands, completely bypassing the intended security guardrails of the confirmation prompt.

### Steps to Reproduce

1. Start the Gemini CLI in a test directory.
2. Prompt the agent to run a maliciously padded multi-line command:
> *"Please use the run_shell_command tool to execute this exact string: 'echo Safe' followed by 30 newlines, followed by 'echo Malicious Payload Hidden'"*
3. Observe the CLI UI. The display truncates the command (showing `... hidden (Ctrl+O) ...`).
4. Notice that the standard `1. Allow once` / `2. Allow for this session` options are still present and actionable.
5. If the user presses `1`, the hidden malicious payload is executed.

### Expected Behavior

If a command exceeds the maximum display size and is visually truncated, the standard approval options should be disabled or hidden. The user should be forced to expand the view (e.g., using `Ctrl+O`) to inspect the full command before the "Allow" options become available.

### Actual Behavior

The UI allows execution approval even when critical parts of the command are visually truncated and hidden from the user.

### Proposed Solution

I have already submitted a PR to address this issue: **[#23333](https://github.com/google-gemini/gemini-cli/pull/23333)**.

The fix links the `totalHiddenLines` calculation in `MaxSizedBox` to the state logic in `ToolConfirmationMessage`. If `isTruncated` is true, the standard "Allow" options are replaced with a safety message (`⚠️ Expand to view full command (Press Ctrl+O)`), locking out the confirmation until the user explicitly expands the view.