Create a chat completion directly against OpenRouter

Proxies the request body to OpenRouter's OpenAI-compatible /chat/completions. Most parameters OpenRouter accepts are forwarded verbatim (sampling knobs, tools, tool_choice, response_format, reasoning, transforms, route). Two are not: provider is merged, not forwarded as-is — caller routing preferences are honoured but the per-token price cap is always applied, so upstream spend never exceeds the rate the call is billed at — and models (OpenRouter's own fallback-routing list) and plugins (paid addons like web search) are stripped and silently ignored: neither is validated against this proxy's catalog/policy/price cap or billed for, so honoring them isn't safe. Responses are returned unchanged apart from an appended openhuman billing envelope. Set stream: true for an SSE stream.