refactor(bot): centralize runtime state; support albums + safer vision handling

- make MessageStore.put() return StoredMessage and allow collectReplyChainText() to work with StoredMessage
- move muted users + answers loading into Environment (add Answers model + GEMINI_IMAGE_MODEL)
- extract message handling into processNewMessage() and add media-group (album) caching/downloading by unique_file_id
- for Ollama: check model capabilities before sending images; use replyToMessage helper consistently
- add /geminiGenImage command stub for Gemini image generation

src/model/answers.ts

@@ -0,0 +1,9 @@
+export type Answers = {
+    test: string[]
+    prefix: string[]
+    better: string[]
+    who: string[]
+    kick: string[]
+    invite: string[]
+    day: number[]
+}