[RyzenAI] Erroneous `vitis_ai_report.json` with RyzenSDK 1.1

Ryzen SDK: 1.1

Upon executing tests, the terminal logs exhibit the following entries:

```bash
[Vitis AI EP] No. of Operators : CPU 2 IPU 334 30.12% CPU 2 IPU 334 30.12% CPU 2 IPU 372 33.54% CPU 2 IPU 372 33.54% CPU 3 IPU 1106 99.73%
[Vitis AI EP] No. of Subgraphs : CPU 1 IPU 1 Actually running on IPU 1 CPU 1 IPU 1 Actually running on IPU 1 CPU 1 IPU 1 Actually running on IPU 1 CPU 1 IPU 1 Actually running on IPU 1 CPU 1 IPU 1 Actually running on IPU 1
```

Similarly, the generated `vitis_ai_report.json` also contains multiple entries for "All", "DPU" and "CPU" operators.

This issue arises when running multiple models in a loop, where it appears that the results of the previous model are unintentionally appended to the current results.

See: https://github.com/huggingface/optimum-amd/actions/runs/8356674296/job/22874227285#step:5:272

Steps to Reproduce:
```
* Run: $env:RUN_SLOW=1; pytest -s -m "prequantized_model_test" .\tests\ryzenai\test_modeling.py
* Review the generated prints and `vitis_ai_report.json` for each model in the `ryzen_cache` folder.Check the generated prints and `vitis_ai_report.json` for each model in `ryzen_cache` folder.
```

Minimal example to test:
```python
import os

import requests
from PIL import Image

from optimum.amd.ryzenai import RyzenAIModelForImageClassification, pipeline
from tests.ryzenai.testing_utils import DEFAULT_VAIP_CONFIG

def test_pipeline(model_id):
os.environ["XLNX_ENABLE_CACHE"] = "0"
os.environ["XLNX_USE_SHARED_CONTEXT"] = "1"

cache_dir = ".image-classification-pipeline-cache"
cache_key = model_id.replace("/", "_").lower()

provider_options = {}
provider_options["cacheDir"] = cache_dir
provider_options["cacheKey"] = cache_key

model_instance = RyzenAIModelForImageClassification.from_pretrained(
model_id, vaip_config=DEFAULT_VAIP_CONFIG, provider_options=provider_options
)

pipe = pipeline("image-classification", model=model_instance, vaip_config=DEFAULT_VAIP_CONFIG)

url = "http://images.cocodataset.org/val2017/000000039769.jpg"
image = Image.open(requests.get(url, stream=True).raw)
pipe(image)[0]

model_ids = ["mohitsha/timm-resnet18-onnx-quantized-ryzen", "mohitsha/transformers-resnet18-onnx-quantized-ryzen"]
for model_id in model_ids:
test_pipeline(model_id)
```

Expected behavior:
* Clean log statements with no carryover of results from previous model executions.