Add new backend support for `TensorRT-LLM`

[wtlow003]

Seen from the benchmark here: https://github.com/shashikg/WhisperS2T?tab=readme-ov-file

We observed that WhisperS2T provides impressive performance when using the backend of TensorRT-LLM.