Proxy embedding requests through the inference gateway

Hive’s model gateway now proxies embedding requests through a new POST /inference/v1/embeddings endpoint. Operators can define an embedding profile, point OpenAI-compatible embedding clients at Hive’s /inference/v1 base address, and authenticate using the Hive profile token. Each request is validated against the token-bound profile, rewritten to the configured upstream embedding model, and forwarded without exposing provider credentials. Requests are recorded as a distinct embedding operation on inference usage rows, with cost estimated from input tokens. This lets centralizing clients route document embeddings through Hive instead of pointing them directly at upstream providers.