Enhance performance of batch inferencing tutorial with vllm and running on L40s and H100 GPUs

[jeremiaswerner]

Following changes have been performed:

• rework the application code to use vLLM
• use the Granite-4.0-Micro model from Huggingface
• store the model on the input store as a model cache to speed up subsequent runs
• support for concurrency on H100 to support larger models

As a result, this tutorial can now support high-throughput batch inferencing use cases with any LLM.

Rendered version:
https://github.com/IBM/CodeEngine/blob/4a58342856c3828b7477f67af9aa2c8338d41a47/serverless-fleets/tutorials/inferencing/README.md

Since I added 8000 recipes pls review at a per commit basis

[reggeenr]

LGTM