Secure credential storage for HTTP MCP OAuth tokens

### Describe the feature or problem you'd like to solve

The Copilot CLI currently persists OAuth access and refresh tokens as plaintext JSON files on disk (~/.copilot/mcp-oauth-config/*.tokens.json) when authenticating to HTTP-based MCP servers. This introduces several security concerns:
1. **Plaintext exposure** - Tokens stored as unencrypted JSON can be read by any process or user with filesystem access. This violates the principle of least privilege for credential storage.
2. **Cloud sync risk** - Many enterprise users sync their home directory or profile folders via OneDrive, Dropbox, or similar tools. Plaintext tokens in synced directories get uploaded to cloud storage, expanding the attack surface and triggering DLP/security policy violations.
3. **Credential persistence** - File-based tokens survive beyond their intended session lifetime. If a machine is compromised, an attacker gains access to cached tokens without needing to re-authenticate.
4. **Inconsistency with VS Code** - VS Code already handles MCP OAuth tokens securely through the OS Secret Storage API (Windows Credential Manager, macOS Keychain, libsecret on Linux). The CLI should follow the same pattern for consistency and security.

### Proposed solution

Leverage the OS-native credential/secret manager on each platform rather than writing tokens to disk. This aligns with security best practices, avoids DLP policy conflicts, and provides a consistent security posture across the Copilot toolchain.

### Example prompts or workflows

_No response_

### Additional context

_No response_

Start by locating the code that writes OAuth tokens to ~/.copilot/mcp-oauth-config/*.tokens.json. Review how the CLI authenticates to HTTP-based MCP servers and how VS Code uses OS Secret Storage, including Windows Credential Manager, macOS Keychain, and libsecret on Linux. Done means tokens no longer persist as plaintext files and are stored through native credential managers on each platform.

Written by the indexing model from the issue text.