feat(agent-server): add non-streaming OpenAI chat completions gateway

[enyst]

Problem

OpenAI-compatible clients need a minimal way to call an OpenHands agent through the agent-server using the de facto /v1/chat/completions and /v1/models surfaces.

Scope

• Add GET /v1/models for profile-backed openhands_{profile} model IDs.
• Add non-streaming POST /v1/chat/completions backed by agent-server conversations.
• Map OpenAI-style Bearer auth to the existing session key auth.
• Support optional X-OpenHands-ServerConversation-ID reuse while keeping tool activity internal.
• Return final assistant text and usage metrics when available.

Existing implementation PR: #3545

Parent: #3540
Source roadmap: #3540 (comment)

This issue was created by an AI agent (OpenHands) on behalf of the user.