[issue-5931] [FE/P SDK] fix: render Google ADK attachments in prompt parts

[ollieagent]

Details

Fixes two layered bugs that prevented Google ADK image attachments from being extracted and rendered correctly.

Python SDK — URL-safe base64 not matched by regex

When a genai_types.Part contains inline_data (a Blob with raw bytes), pydantic's model_dump(mode='json') serializes the bytes as URL-safe base64 (using - and _ instead of + and /). The AttachmentsExtractor regex only matched standard base64 characters, causing it to match only a partial substring — leaving a dangling suffix after the placeholder and producing a broken attachment.

• attachments_extractor.py: extended the base64 character class to include - and _
• decoder_base64.py: added a fallback to base64.urlsafe_b64decode when standard decoding fails, so already-URL-safe strings are decoded correctly

Frontend — Google format handler was null

The google entry in the PrettyLLMMessage format registry was registered as null, so Google ADK traces rendered as raw JSON instead of a structured chat view.

• Added detector.ts: detects Google ADK contents[].parts[] input format and candidates[] output format
• Added mapper.ts: maps parts[] to block descriptors — inline_data → image blocks, text → text blocks, function_call/function_response → code blocks; maps model role → assistant, function role → tool
• Added index.ts: exports the googleFormat implementation
• Registered googleFormat in registry.ts
• PromptMessagesReadonly.tsx: added inline_data handling so the existing prompt message readonly view also renders Google-format images

Change checklist

• User facing
• Documentation update

Issues

• Resolves [Bug]: Google ADK integration failing to render attachments in prompt parts #5931

Testing

1. Create an ADK trace with an image attachment using the Python SDK:

   import base64, opik
   from opik.api_objects.attachment import AttachmentType
   
   client = opik.Opik()
   png_bytes = b'\x89PNG\r\n...'  # any PNG bytes
   b64 = base64.urlsafe_b64encode(png_bytes).decode()
   trace = client.trace(
       name="adk-test",
       input={"contents": [{"role": "user", "parts": [{"text": "Analyze"},
           {"inline_data": {"data": b64, "mime_type": "image/png"}}]}]},
       output={"text": "Done"}
   )
   trace.end()

2. Open the trace in the UI → Messages tab.
3. Verify:
   • The input renders as a User message with a text block and an image attachment block
   • The output renders as an Assistant message
   • No raw JSON is shown

Unit tests:

• Python: pytest tests/unit/api_objects/attachment/test_attachments_extractor.py (31 tests)
• Frontend: npx vitest run src/shared/PrettyLLMMessage (94 tests)

Documentation

No documentation changes required.

Screenshots

[andrescrz]

Please remove these screenshots, these should be comitted.