feat: add Ollama audio transcription and runtime config reload

- add audio capability reporting for Ollama models
- support Telegram voice messages via ffmpeg conversion and Ollama transcription
- add USE_SYSTEM_PROMPT toggle and runtime reloading of .env/system prompt settings
- support ollama_options.json for custom Ollama request options
- improve Telegram MarkdownV2 escaping and formatting preservation
- add environment setters for AI provider credentials and models
- show audio capability in info/model commands

src/commands/mistral-chat.ts

@@ -59,7 +59,7 @@ export class MistralChat extends ChatCommand {
         });
         chatMessages.reverse();
 
-        if (Environment.SYSTEM_PROMPT) {
+        if (Environment.SYSTEM_PROMPT && Environment.USE_SYSTEM_PROMPT) {
             chatMessages.unshift({role: "system", content: [{type: "text", text: Environment.SYSTEM_PROMPT}]});
         }