fix(analyze): use AI SDK ImagePart format for vision screenshots (#5566)

The analyze plugin used OpenAI's `image_url` format for screenshot
attachments, but `generateText()` from the Vercel AI SDK expects
the `ImagePart` format (`type: 'image'`). This caused "Invalid prompt:
The messages do not match the ModelMessage[] schema" when vision: true.

Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>

Author: gololdf1sh

lib/plugin/analyze.js

@@ -155,10 +155,9 @@ const defaultConfig = {
       if (config.vision && test.artifacts.screenshot) {
         debug('Adding screenshot to prompt')
         messages[0].content.push({
-          type: 'image_url',
-          image_url: {
-            url: 'data:image/png;base64,' + base64EncodeFile(test.artifacts.screenshot),
-          },
+          type: 'image',
+          image: base64EncodeFile(test.artifacts.screenshot),
+          mediaType: 'image/png',
         })
       }