[tests] [TRTLLM-3602][feat] support nvfp4 model and fp8 kv cache for MLA chunked prefill (Blackwell) (#5475) (8836990)

## Commit Details for tests Directory Update

**Author:** jmydurant
**Date:** Thu Jun 26 22:18:08 2025 +0800
Please review these changes and decide if you need to make corresponding updates in your implementation.
[View original commit](https://github.com/NVIDIA/TensorRT-LLM/commit/8836990bde20a4d3fc64a2aabe78b427e6899120)