feat(ai): 🤖 update LLM models and configuration

- Changed default model from `ollama:llama3.2:3b` to `ollama:gemma3:4b`
- Updated model token limits and added new models:
  - Added `gemma3:4b` with vision support
  - Removed `phi4:14b` and added `phi4-reasoning:14b`
  - Updated token limits for existing models
- Added new AI-related roles:
  - `commit-message.md`
  - `email-answer.md`
  - `emoji-commit-message.md`
  - `git-branch.md`
  - `improve-prompt.md`
  - `improve-writing.md`
  - `linkedin-answer.md`
  - Enhanced `to-emoji.md` and added `to-emojies.md`
- Added `ai` alias for `aichat` in fish config
- Updated NixOS configuration to load new models

home/.config/aichat/roles/to-emoji.md

@@ -1,6 +1,22 @@
 ---
-model: ollama:llama3.2:3b
+model: ollama:gemma3:4b
 
 ---
-convert __INPUT__ to emoji.
-response: just an emoji or emojies, without words or explanations
+# Emoji Conversion Prompt
+
+**Task:**  
+Convert the user's input text into a single, most representative emoji that captures the core meaning or emotion of the message.  
+
+**Rules:**  
+1. Respond **only** with one emoji—no words, explanations, or additional characters.  
+2. Choose the emoji that best symbolizes the essence of the input (e.g., "happy" → 😊, "rain" → 🌧️).  
+3. If the input is ambiguous or unclear, default to a neutral symbol (e.g., ❓ or 🤔).  
+
+**Example Inputs & Outputs:**  
+- "I love pizza" → 🍕  
+- "I'm so tired" → 😴  
+- "Celebration!" → 🎉  
+
+**Note:**  
+- Avoid combining multiple emojis.  
+- Prioritize universally recognized emojis (Unicode standard).