fix(vllm): structured outputs silently ignored on vLLM >= 0.23 (GuidedDecodingParams removed)

[pos-ei-don]

Supersedes #10339, which got auto-closed when I force-pushed from a shallow clone that had severed the commit's parent (the diff briefly blew up to the whole tree and GitHub closed it — sorry for the noise). This branch is now cleanly based on current master with a single 6-line change.

Description

On vLLM >= 0.23, GuidedDecodingParams was removed from vllm.sampling_params (replaced by StructuredOutputsParams, and the SamplingParams field renamed guided_decoding -> structured_outputs). The vLLM backend's import therefore fails:

try:
    from vllm.sampling_params import GuidedDecodingParams
    HAS_GUIDED_DECODING = True
except ImportError:
    HAS_GUIDED_DECODING = False

HAS_GUIDED_DECODING becomes False, so the entire guided-decoding block is skipped and response_format / grammar constraints are silently ignored — the model returns unconstrained text.

Reproduce (vLLM 0.23.0, request with "response_format": {"type": "json_schema", ...}):

• Before: plain text (e.g. Innsbruck), schema not enforced.
• After: schema-conforming JSON (e.g. {"country":"Österreich","name":"Innsbruck","population":132493}).

Verified on NVIDIA GB10 / arm64 / CUDA 13, vLLM 0.23.0.

Fix

Adapt the existing request.Grammar path to the new class/field (StructuredOutputsParams / structured_outputs). Per @richiejp's review on #10339, no backwards-compat branch — the backend tracks the latest vLLM, so there's no supported version below 0.23 to fall back to.

[pos-ei-don]

Heads-up: the tests failure here is unrelated to this change. TestAllFieldsHaveRegistryEntries (core/config/meta/registry_coverage_test.go) fails on the config field pipeline.max_history_items (added in #10331), which has no entry in core/config/meta/registry.go and isn't grandfathered — so the coverage gate trips on current master and any PR branched off it.

All functional specs pass (168/168 unit + 92/101 E2E, 9 skipped). This diff is 6 lines in backend/python/vllm/backend.py and can't affect a Go config-registry gate. Happy to send a separate one-line PR adding the registry entry if that helps.