[FEATURE]: Convert A2UI express compiler tests to conformance tests

The A2UI express compiler/decompiler etc in Python has lots of tests, but they're not platform-agnostic conformance tests.

We want them to be platform agnostic, so we can write equivalent compiler/decompilers in Typescript for example.

Properties we want:

1. Inference format conformance tests should have their own schema for YAML definitions, to avoid needing to make a huge YAML file
2. There should be format-independent tests which do round tripping. These may already exist.
3. There should be format-specific tests which have golden input and output, to test compilation, decompilation and prompt generation.
4. The format-specific tests should be in subfolders e.g. inference_formats/express etc.
5. Initially, wire up the Python agent SDK to have a test harness which runs these tests. Follow the pattern of the existing conformance tests as much as you can use relevant docs and skills.
6. Let's replace programmatic tests where it makes sense. Most tests should be data-driven conformance tests.