Add an OpenAI-compatible /v1/chat/completions endpoint

The server currently exposes a custom `/generate`. An OpenAI-compatible endpoint would let existing clients point at pooji-vllm without changes. **Where:** `pooji_vllm/server.py`. **Scope:** `/v1/chat/completions` and `/v1/models`, with streaming in the OpenAI SSE delta format, chat-template application via `tokenizer.apply_chat_template`, and the usual sampling parameters mapped onto `SamplingParams`. **Good first issue** — self-contained, no engine changes required.