Python: Drop post-limit streamed function calls

[scarab-systems]

Motivation & Context

Fixes #7045.

When max_function_calls is reached, Agent Framework asks the provider for a final response with tool_choice="none". If a streaming provider still returns a function call on that final turn, forwarding it exposes a tool call that the framework will not execute, so transports such as AG-UI can be left waiting for a matching tool result.

Disclosure: This PR was prepared with AI assistance under human direction and review.

Description & Review Guide

• What are the major changes?
  • Drop function_call content from post-limit streaming updates before forwarding them to callers.
  • Strip function_call content from limit fallback responses before deciding whether fallback text is needed.
  • Add regression coverage for a streaming provider returning another function_call after max_function_calls is exhausted.
• What is the impact of these changes?
  • Clients no longer receive a post-limit function call that Agent Framework will not execute, while the existing fallback text behavior remains unchanged when no visible final content is available.
• What do you want reviewers to focus on?
  • Whether dropping the disobedient post-limit call is the right scope for the max-function-call limit path, compared with executing an over-limit call or emitting a synthetic tool result.

Related Issue

Fixes #7045

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 (see Related Issue above).
• This is not a breaking change. If it is a breaking change, add the breaking change label (or add "[BREAKING]" to the title prefix, before or after any language prefix) -- a workflow keeps the label and title prefix in sync automatically.

[Copilot]

Pull request overview

This PR fixes a streaming edge case in the Python Agent Framework function-invocation limit path: when max_function_calls is reached and the framework requests a final tool_choice="none" response, any provider that still streams function_call content could cause clients (e.g., AG-UI) to observe tool-call starts without a corresponding tool result.

Changes:

• Strip function_call content from post-limit streaming updates before forwarding them to callers.
• Strip function_call content from finalized responses in the max-function-calls fallback path before deciding whether to synthesize fallback text.
• Add a regression test covering a streaming provider emitting another function_call after max_function_calls is exhausted.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 1 comment.

File	Description
python/packages/core/agent_framework/_tools.py	Filters post-limit streamed tool-call content and ensures finalized responses don’t retain over-limit function_call items.
python/packages/core/tests/core/test_function_invocation_logic.py	Adds coverage to ensure post-limit streamed function_call updates are not forwarded and no orphaned tool calls appear.

[eavanvalkenburg]

@scarab-systems, thank you for identifying and fixing the post-limit streaming orphan. The consolidated draft PR #7345 incorporates that approach and extends it across both max_function_calls and max_iterations, including native approval requests while preserving metadata-only updates. Closing this PR in favor of the consolidated work.

[scarab-systems]

Woohoo! Thank you for the clear update, and for carrying the fix forward into the consolidated PR. I appreciate the broader treatment across max_function_calls, max_iterations, approval requests, and metadata-only stream updates. Glad the original patch helped identify the right direction.