Python: fix(foundry_hosting): preserve auth credentials across FoundryToolbox reconnections

[PratikWayase]

PR Description

Motivation & Context

When using the ghmcpoauth authentication method with a Foundry Toolbox, accepting an OAuth consent request triggers a close() on the underlying MCP tool. Currently, this permanently nullifies the authenticated _httpx_client. As a result, any follow-up requests in the same session fail with a server_error because the toolbox falls back to an unauthenticated transport. Users currently have to start a completely new conversation to recover from this state. This PR fixes that lifecycle bug so sessions remain valid after OAuth consent flows.

Related Issue

Fixes #7166

Description & Review Guide

• What are the major changes?

  • Modified FoundryToolbox.__init__ to cache the credential, token_scope, and timeout as instance attributes.
  • Overrode get_mcp_client() to lazily recreate an authenticated httpx.AsyncClient if the previous one was closed.
  • Updated the close() docstring to clarify that credentials are preserved for reconnection.
  • Added regression tests in test_toolbox.py to verify credential preservation, idempotent closes, and proper context manager recreation.

• What is the impact of these changes?
  Hosted agents using OAuth2-authenticated toolboxes will no longer crash mid-session after a user completes a consent flow. The fix is entirely internal to the FoundryToolbox lifecycle and does not alter the public API or affect non-OAuth authentication methods.

• What do you want reviewers to focus on?
  Please review the lazy recreation logic in get_mcp_client(). I opted to rebuild the httpx.AsyncClient only when _httpx_client is None rather than trying to resurrect a closed client. I'd appreciate feedback on whether this approach aligns with the framework's expected lifecycle patterns, and whether the new regression tests adequately cover edge cases around repeated close/reconnect cycles.

Contribution Checklist

• The code builds clean without any errors or warnings
• All unit tests pass, and I have added new tests where possible
• The PR follows the Contribution Guidelines
• This PR is linked to an issue and there is no other open PR for this issue
• This is not a breaking change.

[Copilot]

Pull request overview

Fixes the FoundryToolbox lifecycle so OAuth2-authenticated sessions can recover after an internal close() (e.g., during OAuth consent flows) by preserving credentials and recreating an authenticated httpx.AsyncClient on demand.

Changes:

• Cache credential, token_scope, and timeout on the FoundryToolbox instance for reuse after teardown.
• Override get_mcp_client() to lazily recreate an authenticated httpx.AsyncClient when the prior client has been cleared.
• Add regression tests validating credential preservation and reconnection behavior across repeated close() calls.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 2 comments.

File	Description
python/packages/foundry_hosting/agent_framework_foundry_hosting/_toolbox.py	Preserve auth-related state and recreate an authenticated HTTP client on reconnection.
python/packages/foundry_hosting/tests/test_toolbox.py	Add regression tests covering close/reconnect cycles and idempotent close behavior.