test: Add test and usage example for mm requests

[sagearc]

review fixes from #428

Added the example from #280 and mm test

Closes #451, Closes #452

[Copilot]

Pull request overview

Adds multimodal (image) coverage to the UDS tokenizer renderer integration tests by introducing a local image fixture and new MM-specific test cases.

Changes:

• Add local llm-d logo testdata and a fixture that exposes it as a base64 data URL.
• Introduce MM model fixture and new RenderChatCompletion MM integration tests.
• Add helper for building image+text chat messages for test requests.

Reviewed changes

Copilot reviewed 2 out of 3 changed files in this pull request and generated 5 comments.

File	Description
services/uds_tokenizer/tests/testdata/llmd_logo.png	Adds a local image asset used by MM tests.
services/uds_tokenizer/tests/test_renderer.py	Adds MM request helpers and new MM integration tests (features + determinism).
services/uds_tokenizer/tests/conftest.py	Adds MM model fixture and base64 data URL fixture backed by local testdata.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[sagearc]

@delavet CR fixes

closes #452, #451

[delavet]

I think we also need a Dockerfile and corresponding Make commands to build this example—what do you think?
Everything else is fantastic!

[gyliu513]

@delavet Introducing a dockerfile maybe complex to run this program, how about follow other demos like make run-example kv_cache_index, it is just calling go build and then run the binary.

[delavet]

@delavet Introducing a dockerfile maybe complex to run this program, how about follow other demos like make run-example kv_cache_index, it is just calling go build and then run the binary.

Perhaps you’re right—this would make verification simpler. However, I usually do verify in Kubernetes environments, so I may consider submitting a PR to include Kubernetes deployment-related content. Thank you for your feedback!

I'll approve this for now. @sagearc What do you think? Perhaps we can merge this PR now?

[sagearc]

Thanks @delavet @gyliu513, good points. Since examples are mostly meant to run locally, we could add a run-example-online-uds Make target that sets up the local venv and starts the UDS tokenizer as a background process alongside the go binary.
Happy to split if you'd prefer to land the test first.

[gyliu513]

@sagearc +1 to land the test first, we can always have follow up PRs.

[delavet]

@sagearc Thanks, I think we can merge this right after conflicts being resolved

[sagearc]

Opened this PR to update all examples to use the tokenization service:
#471