xgrammar 0.2.3 (PR mlc-ai/xgrammar#668) added an `any_order` field to the
json_schema structural-tag format, so `StructuralTag.model_dump()` now emits
`"any_order": False` inside each tag's `content`. This broke the hardcoded
expected dict in `test_get_model_structural_tag_supports_vllm_hermes`, which
runs only in the CPU job (the sole lane that executes `tests/tool_parsers/`
and freshly resolves the open xgrammar range when building its image).
Adopt 0.2.3 rather than pinning below it:
- bump the floor to `xgrammar >= 0.2.3` in requirements/common.txt
- sync the ROCm test lockfile pin to 0.2.3
- add `"any_order": False` to the expected content dicts in the test
Verified against xgrammar 0.2.3 that `JSONSchemaFormat.model_dump()` emits
`any_order` as a sibling of `style`, matching the updated expectations.
Co-authored-by: Claude <noreply@anthropic.com>
Signed-off-by: Kevin Luu <kevin@inferact.ai>
Signed-off-by: khluu <khluu000@gmail.com>
