Add model / test for Qwen3 ASR and example for more granular test outputs

[telnetdoogie]

Just because.
I assumed the Audio file was there for a reason :)
I can't get the whisper test working while keeping everything else working too.

[telnetdoogie]

Also split out the existing VLM integration test in a similar way using a fixture and Results dataclass, perhaps as a template for future tests.

When using a text fixture like this and individual tests for each assertion, test failures will be more descriptive, and we can observe unique differences between test runs versus just "this failed"

[telnetdoogie]

Converting to draft for now. With the uv.lock and pyproject.toml update I'd like to spend some time playing with the resulting build and make sure things are looking good beyond the tests.

I'd propose that as we discover version-compatibility issues or challenges, we can attempt to make a failing test against that configuration for the specific issue. this will help to build a strong corpus of tests that will more affirmatively validate that version updates still work ok.

Turns out that that whisper integration test was actually a very good one to expose a version conflict problem that WAS actually resolvable without breaking other things.

[telnetdoogie]

I've tested this and have found no issues, however maybe someone else might want to run through some of their own runtime / real world tests to make sure things are still working OK for them with the updated dependencies.

The tests don't present an issue, however the updated uv.lock and pyproject.toml which make the tests all run green make me somewhat nervous just given my 'apprentice' skillset with this app.