[ROCM ep][fp16]: when to run stable-diffusion-xl, assert on rocm ep

@cloudhan is already working on this.

Since Jan 27, 2024.

### Describe the issue

assert info: "MultiHeadAttention of ROCm EP is only supported if tunable op is used and tuning is enabled".

### To reproduce

export onnx model as [the official guide](https://github.com/microsoft/onnxruntime/tree/main/onnxruntime/python/tools/transformers/models/stable_diffusion ):
`pip install optimum diffusers onnx onnxruntime-gpu
optimum-cli export onnx --model stabilityai/stable-diffusion-xl-base-1.0 --task stable-diffusion-xl ./sd_xl_base_onnx`
`python optimize_pipeline.py -i ./sd_xl_base_onnx -o ./sd_xl_base_fp16 --float16`

and then, run as:
`from optimum.onnxruntime import ORTStableDiffusionXLPipeline

pipeline = ORTStableDiffusionXLPipeline.from_pretrained('sd_xl_base_fp16', provider='ROCMExecutionProvider')
prompt = "a cat, cinematic photo cute pet . 35mm photograph, film, bokeh, professional, 4k, highly detailed"
image = pipeline(prompt=prompt, num_images_per_prompt=1, num_inference_steps=50, guidance_scale=0).images[0]

image.save("output_fp16.png")`

### Urgency

_No response_

### Platform

Linux

### OS Version

ubuntu18.04

### ONNX Runtime Installation

Built from Source

### ONNX Runtime Version or Commit ID

2b87dd373a3567c2c426e2f090b201b8b051a346

### ONNX Runtime API

Python

### Architecture

X64

### Execution Provider

Other / Unknown

### Execution Provider Library Version

rocm5.2