Enable nothink models - silent responses for Qwen3 nothink models under default thinking config.

[gherlein]

Problem

Qwen3 -nothink model variants (e.g. qwen3-coder-next-nothink:latest) produced
no visible response in the TUI. Two bugs combined to cause this:

1. The default thinkingLevel: "medium" config caused every Ollama chat request to
   include think: "medium", which overrides the /no_think token baked into
   nothink model templates — forcing the model into thinking mode.

2. In ollamaRunStreaming, when a model responds entirely via thinking tokens
   response was built from aggregatedText only. The thinking content was silently
   discarded, producing an empty turn — "prompts do nothing".

Fix

GenerateContent — detect nothink in the model name and explicitly set
think: false, preventing the configured thinking level from overriding the
model's intent.

ollamaRunStreaming — when aggregatedText is empty but aggregatedThinking
is not, fall back to surfacing the thinking content as the response. This mirrors
the existing behaviour in ollamaRunNonStreaming and acts as a safety net for any
model that ends up in this state.

Testing

Manually verified with qwen3-coder-next-nothink:latest (previously silent) and
qwen3.5:35b-a3b-q4_k_m (working model, confirmed no regression).