Streaming parser does not handle Ollama's reasoning field for thinking models #805

Description

When using thinking/reasoning models via Ollama (e.g. Qwen 3.5), Ollama's OpenAI-compatible streaming endpoint places the model's thinking output in a field named reasoning within the delta object. The streaming parser in openai.rs (around line 1152) only checks for reasoning_content, not reasoning. As a result, the thinking output is silently dropped.

Might be similar to issue #785

Ollama streaming chunk example:
{"choices":[{"delta":{"role":"assistant","content":"","reasoning":"Thinking Process:\n..."}}]}

The parser checks:

• delta["content"] — empty for thinking models
• delta["reasoning_content"] — not present in Ollama's response

It does not check delta["reasoning"], which is where Ollama places the output.

Expected Behavior

The parser should also check delta["reasoning"] as a fallback, e.g.:
if let Some(reasoning) = delta["reasoning_content"].as_str().or_else(|| delta["reasoning"].as_str()) {

Steps to Reproduce

1. Pull a thinking model in Ollama: ollama pull qwen3.5
2. Configure OpenFang with Ollama and qwen3.5:latest
3. Send a chat message
4. The thinking output is dropped — no text appears

OpenFang Version

0.5.1

Operating System

Linux (x86_64)

Logs / Screenshots

Version: 0.5.1 (confirmed by reading source at crates/openfang-runtime/src/drivers/openai.rs)