Problem

When users execute /reasoning off or /think off in Telegram, reasoning content
is still sent as separate messages. This happens because splitTextIntoLaneSegments()
extracts reasoning text from model output without checking if reasoning is disabled.

This is the Telegram-specific follow-up to issue #9563, which was previously fixed
for Discord in commit 14c54e6.

Root Cause

The splitTextIntoLaneSegments() function in bot-message-dispatch.ts always
creates reasoning segments when <think> tags or "Reasoning:" prefixes are found
in the model output. It doesn't respect the user's reasoningLevel setting.

// BEFORE: Always adds reasoning segments if text contains reasoning
if (split.reasoningText) {
  segments.push({ lane: "reasoning", text: split.reasoningText });
}

Solution

Filter out reasoning segments when resolvedReasoningLevel === "off". The
resolvedReasoningLevel variable is already available in scope (defined at line 144).

// AFTER: Only adds reasoning segments if reasoning is not explicitly disabled
if (split.reasoningText && resolvedReasoningLevel !== "off") {
  segments.push({ lane: "reasoning", text: split.reasoningText });
}

Changes

• src/telegram/bot-message-dispatch.ts: Add resolvedReasoningLevel !== "off"
  condition to the reasoning segment check (1 line change)

Testing

• Manual: Start conversation with reasoning-capable model (e.g., Kimi K2.5)
• Run /reasoning off
• Send a message that would normally trigger reasoning
• Verify no separate "Reasoning:" message appears in Telegram
• Verify answer content still appears correctly

Related Issues

• Fixes Fix: /think off does not disable reasoning_content for reasoning-capable models #9563 (Telegram-specific portion)
• Related: 14c54e6 (Discord fix for same issue)

Checklist

• Surgical fix (minimal code change)
• No breaking changes
• Follows existing code patterns
• Uses already-available variable in scope