fix: use provider block's own vision model on legacy fallback

When falling back to a provider block (providerID / legacy ludmila-ai),
prefer the model flagged attachment:true in that block instead of the
Google-default options.model. Prevents sending 'gemini-2.5-flash' to the
ludmila endpoint which expects 'gemini/gemini-2.5-flash'.

package.json

@@ -1,6 +1,6 @@
 {
   "name": "vision-bridge",
-  "version": "0.2.0",
+  "version": "0.2.1",
   "type": "module",
   "description": "Opencode plugin: bridges images to a vision model when the active model cannot see images.",
   "main": "src/vision-bridge.ts",