OpenAI-compatible inference relay with model-bound profiles and usage accounting

Hive now ships as an OpenAI-compatible inference relay. Profiles bind a stable model name to an upstream provider (configured in the dashboard or via environment), and tokens authenticate and attribute requests to a profile. Relay requests record input and output token counts as well as an estimated US dollar cost, and the operations dashboard adds dedicated Inference views for Profiles, Providers, profile and token detail pages, usage charts and widgets, plus token and provider creation flows. Models.dev is used to validate model identifiers. Streamed chat completions are supported, and existing agent settings are unaffected.