fix: use ReasoningContent not Content for pre-tool thinking display

The pre-tool IterationCallback was passing result.Content (visible assistant
text) instead of result.ReasoningContent (DeepSeek thinking tokens). This
meant the 💭 reasoning messages only appeared when the model also output
visible text alongside tool calls — which DeepSeek rarely does.

Fixes the Telegram bot not showing deterministic progress / thinking notes.

Author: molty3000

internal/loop/loop.go

@@ -602,7 +602,7 @@ func (e *Engine) runLoop(ctx context.Context, messages []llm.Message) (string, [
 				CachedTokens:        e.TotalCachedTokens,
 				TotalLatency:        time.Since(startTime),
 				HasFinalAnswer:      false,
-				ReasoningContent:    result.Content,
+				ReasoningContent:    result.ReasoningContent,
 				IsPreTool:           true,
 			})
 		}