MLC LLM README (read 2026-10-04): negative result on the feature axes

Read 2026-10-04 specifically to fill the feature-comparison axes for [[mlc-llm]], which had none of the seven axes recorded. THIS IS A NEGATIVE SOURCE AND IS RECORDED AS ONE: a case-insensitive grep across the README for speculat|eagle|mtp|grammar|json.schema|structured|chunked|prefix cach|radix|disagg|multi.model|multimodal|vision|audio returned ZERO matches. The README is a short project overview (install, supported hardware, model library) and simply does not discuss these features, which is NOT the same as the engine lacking them - the model library at https://mlc-ai.github.io/mlc-llm/models is the authoritative per-model surface and was not enumerated here. So for [[mlc-llm]]: speculation, chunked prefill, disaggregation and structured output are UNSTATED-AND-UNVERIFIED from this source, prefix caching is UNSTATED even though the record asserts it (the assertion is not contradicted, only unconfirmed here), and the multi-model claim is inherited from the record's existing 'One MLCEngine behind Python, REST, JavaScript/WebGPU, iOS and Android' bullet which is about SURFACES, not about serving several models at once - a distinction this source makes worth recording.