Problem

When using a third-party OpenAI-compatible API provider that returns reasoning_content fields in streaming SSE chunks, deepseek-tui incorrectly treats every token as a separate "thinking" block instead of normal text output. This causes severe display corruption where each character/token is rendered with its own thinking done · 0.0s header.

Fixes #1673

Root Cause

1. is_reasoning_model() checks only the model name (e.g., deepseek-v4-pro, deepseek-v4-flash) to decide whether reasoning_content deltas should be treated as thinking blocks — it does not consider whether the current provider actually supports reasoning content semantics.

2. When a non-DeepSeek provider streams reasoning_content fields alongside normal response text, every chunk triggers the thinking UI path, corrupting the display.

Fix

• Added requires_reasoning_content() helper that returns true only for the native DeepSeek provider (ProviderKind::Deepseek).
• Updated handle_chat_completion_stream() to use requires_reasoning_content(api_provider) instead of is_reasoning_model(model) when deciding whether to emit StreamEvent::ReasoningDelta.
• Updated parse_sse_chunk() to accept the provider kind and use the same gate.

This ensures third-party OpenAI-compatible providers that happen to serve DeepSeek models are not incorrectly treated as supporting DeepSeek's reasoning content protocol, while preserving full native DeepSeek reasoning display.