[DO NOT MERGE yet] Update wordle tutorial to native vllm serve

[sergiopaniego]

What

Updates the vLLM server command in tutorial/examples/wordle.py from the deprecated trl vllm-serve to vLLM's native server, matching the migration in huggingface/trl#6765:

CUDA_VISIBLE_DEVICES=0 VLLM_SERVER_DEV_MODE=1 vllm serve Qwen/Qwen3-1.7B --host 0.0.0.0 --port 8000 \
    --weight-transfer-config '{"backend": "nccl"}' \
    --logprobs-mode processed_logprobs \
    --max-logprobs -1

Do not merge yet

The tutorial's uv header installs trl[vllm] from PyPI. The released TRL client (v1.10.0) still speaks the old TRL server protocol, so this command only works once:

1. Use vLLM's own server instead of TRL's trl#6765 is merged, and
2. a TRL release containing it is out.

Until then, the current trl vllm-serve command in the tutorial remains the correct one (and it keeps working after the release too, as a deprecated wrapper until TRL v2.0). Will mark ready for review once the TRL release ships.

AI-assisted PR.