Feature: Add reasoning event support to Gemini CLI bridge

[onmete]

🚀 Feature Description

Is your feature request related to a problem?
The ag_ui_gemini_cli adapter (ag_ui_gemini_cli/adapter.py) only emits TEXT_MESSAGE_*, TOOL_CALL_*, RUN_STARTED/FINISHED/ERROR, and MESSAGES_SNAPSHOT events. There are no REASONING_MESSAGE_START/CONTENT/END events anywhere in the Gemini bridge code. This means the Gemini models' reasoning/thinking output is silently discarded.

Describe the solution you'd like
Add support for emitting REASONING_MESSAGE_START, REASONING_MESSAGE_CONTENT, and REASONING_MESSAGE_END AG-UI events in the Gemini CLI adapter, similar to how other bridges handle reasoning output.

💡 Proposed Solution

Detailed Description: Parse the Gemini CLI's reasoning/thinking output from the NDJSON stream and emit the corresponding AG-UI reasoning events so that reasoning traces are surfaced to the frontend.

User Experience: Users will be able to see the model's reasoning/thinking process in the UI, providing better transparency into how the model arrives at its answers.

🎯 Use Cases

Primary Use Case: Developers and users interacting with Gemini models through the Ambient platform who want visibility into the model's reasoning process.

User Stories:

• As a user, I want to see the Gemini model's reasoning process so that I can understand how it arrived at its answer
• As a developer, I want reasoning events from Gemini so that the UI can display thinking traces consistently across all model providers

🔧 Technical Considerations

Component: Ambient Agentic Runner — Gemini CLI bridge

File: components/runners/ambient-runner/ag_ui_gemini_cli/adapter.py

Implementation Approach: Identify the Gemini CLI's reasoning/thinking output format in the NDJSON stream, then map those entries to REASONING_MESSAGE_START/CONTENT/END AG-UI events in the adapter's event processing loop.

Breaking Changes: No — this is additive. Existing events are unaffected.

✅ Acceptance Criteria

• Gemini CLI reasoning/thinking output is parsed from the NDJSON stream
• REASONING_MESSAGE_START, REASONING_MESSAGE_CONTENT, and REASONING_MESSAGE_END events are emitted
• Reasoning traces appear in the frontend UI when using Gemini models
• Existing event types remain unaffected
• Tests added for reasoning event emission

🏷️ Labels

• Priority: medium
• Effort: M
• Component: backend
• Type: enhancement

[Gkrumbach07]

AI triage: Created session triage-gh-1127-reasoning-events to investigate and fix this issue.