Python: Fix: coalesce reasoning deltas into single block when content.id is None

[PratikWayase]

Motivation & Context

1. Why is this change required? When a provider (like Ollama) streams model reasoning ("thinking"), it often does not provide a stable content.id per chunk, resulting in content.id=None for every delta.
2. What problem does it solve? The AG-UI reasoning emitter (_emit_text_reasoning) was calling generate_event_id() on every single delta when content.id was None. This caused the stream to emit a REASONING_START -> CONTENT -> REASONING_END lifecycle per token, instead of one lifecycle per turn.
3. What scenario does it contribute to? Any AG-UI frontend (e.g., CopilotKit) connected to a "thinking" model (like DeepSeek-R1 via Ollama) was rendering dozens of fragmented "Thought for a few seconds" accordions, with sentences chopped mid-word across them.

Description & Review Guide

• What are the major changes?
  Changed the message_id assignment logic in _emit_text_reasoning() (_run_common.py). It now reuses the existing flow.reasoning_message_id when content.id is None, falling back to generate_event_id() only on the very first delta of a turn. This brings it into parity with how _emit_text() already handles text deltas via flow.message_id. Added 3 regression tests.
• What is the impact of these changes?
  Frontends will now correctly receive a single REASONING_START -> (many) REASONING_MESSAGE_CONTENT -> REASONING_END block per turn. This is a non-breaking change; providers that do supply a content.id continue to use it exactly as before.
• What do you want reviewers to focus on?
  Please verify the 5-line logic swap in _emit_text_reasoning and the 3 new tests in TestReasoningCoalescing to ensure the coalescing behavior correctly mirrors _emit_text.

Remaining Work (Out of Scope for this PR)

The original issue report contained a second, distinct bug regarding parallel tool calls. When Ollama issues the same tool twice in one turn, agent-framework-ollama's _parse_tool_calls_from_ollama reuses the function name as the call_id, causing a collision. That bug is located in agent-framework-ollama and will be addressed in a separate PR.

Related Issue

Partially addresses #6788

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).

[Copilot]

Pull request overview

This PR fixes AG-UI reasoning streaming so that providers which emit reasoning deltas with content.id=None (e.g., Ollama “thinking”) produce a single REASONING_START → (many) REASONING_MESSAGE_CONTENT → REASONING_END lifecycle per turn, instead of fragmenting into a new lifecycle per delta.

Changes:

• Updated _emit_text_reasoning() to reuse flow.reasoning_message_id when content.id is absent, only generating a new ID on the first delta of a reasoning block.
• Added regression tests validating coalescing behavior, explicit content.id behavior, and “new turn/new id” behavior after _close_reasoning_block().

Reviewed changes

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

File	Description
python/packages/ag-ui/agent_framework_ag_ui/_run_common.py	Reuses flow.reasoning_message_id for content.id=None reasoning deltas to prevent per-delta reasoning block fragmentation.
python/packages/ag-ui/tests/ag_ui/test_run_common.py	Adds regression coverage ensuring reasoning deltas coalesce into a single block/message_id per turn and reset correctly across turns.

[antsok]

Fixes #6787