Initial release of the vLLM plugin as a standalone package built on the formae
Plugin SDK. Declaratively manage LoRA adapters on a running vLLM server and
discover the base models it serves, with offline-aware reconciliation for edge
and air-gapped inference fleets.