Python: fix(openai): guard against null delta in streaming chunks from non-co…

[Serjbory]

Motivation and Context

Fixes #5732.

When using OpenAIChatCompletionClient with streaming, certain OpenAI-compatible providers (e.g. Azure OpenAI with specific configurations) send "delta": null on finish chunks instead of the spec-compliant "delta": {}. This causes an AttributeError: 'NoneType' object has no attribute 'content' crash in _parse_text_from_openai, terminating the stream mid-response.

Description

Added a runtime guard in _parse_response_update_from_openai that checks for a None delta before attempting to parse text, tool calls, or reasoning content from the streaming chunk.

Key design decisions:

• The guard is placed at the call site (inside the for choice in chunk.choices loop) rather than in individual parsing methods, so all downstream content-parsing is protected by a single check.
• finish_reason is extracted before the guard, ensuring terminal stream state is always captured even from non-compliant chunks.
• Uses getattr(choice, "delta", None) instead of choice.delta to avoid a Pyright reportUnnecessaryComparison error, since the OpenAI SDK types delta as non-optional despite it being None at runtime from non-compliant providers.
• reasoning_details access was updated to use the local delta variable instead of re-accessing choice.delta.

Added 5 regression tests covering:

• Core crash scenario (delta=None with finish_reason="stop")
• finish_reason preservation across different stop reasons ("length")
• Spec-compliant empty delta (ChoiceDelta()) is not skipped
• Usage data coexisting with a null delta
• Tool-call parsing skipped when delta is None

Contribution Checklist

• The code builds clean without any errors or warnings
• The PR follows the Contribution Guidelines
• All unit tests pass, and I have added new tests where possible
• Is this a breaking change? If yes, add "[BREAKING]" prefix to the title of the PR.

[Copilot]

Pull request overview

Fixes a streaming crash in the Python OpenAIChatCompletionClient when OpenAI-compatible providers emit non-spec finish chunks with "delta": null, by adding a single guard in the streaming chunk parser and covering the scenario with regression tests.

Changes:

• Add a delta is None guard in _parse_response_update_from_openai to prevent downstream parsing from dereferencing None while still capturing finish_reason.
• Update reasoning-details parsing to use the guarded local delta reference.
• Add regression tests for null-delta finish chunks (including finish reason preservation, usage coexistence, and tool-call/text parsing behavior).

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated no comments.

File	Description
python/packages/openai/agent_framework_openai/_chat_completion_client.py	Adds a streaming-time delta=None guard and adjusts reasoning-details access to avoid AttributeError on non-compliant provider chunks.
python/packages/openai/tests/openai/test_openai_chat_completion_client.py	Adds regression tests reproducing and preventing the null-delta streaming crash, including finish-reason and usage behaviors.